2025 AI Security and Incident Review
Key Points
- The episode reviews the past year’s cyber‑security landscape, featuring three segments on AI & data security, incident response, and broader 2025 trends with expert panelists.
- Discussions highlighted the rise of AI‑powered threats, including proliferating AI agents, “shadow AI,” and the need to both defend against AI attacks and protect AI systems from manipulation.
- Specific topics covered ranged from “vibe coding” and cyber‑security sensationalism to emerging quantum threats and insights from IBM’s “Cost of a Data Breach” report.
- The incident‑response segment recapped notable attacks such as the Tea app breach, ClickFix exploit, and various worm campaigns, illustrating the diverse tactics cyber criminals employed over the year.
Sections
- Year-in-Review Cybersecurity Podcast - The hosts introduce a special IBM Security Intelligence episode that recaps the past year’s AI, data security, incident response, and 2025 trends with expert panelists.
- Building Trust in AI Security - The speakers stress the need for observable, resilient AI systems and emphasize trust as the central focus for upcoming AI and data security efforts.
- Securing AI Against Agent Attacks - The speaker warns that as AI enables end‑to‑end attack tools, we must also focus on defending the AI systems themselves—citing Palo Alto’s discovery of agent‑to‑agent session‑smuggling vulnerabilities—and asks if the industry is giving AI protection sufficient attention.
- Risks of Poisoned AI Code Generators - The speakers argue that as AI takes over more coding tasks, attackers could subvert the models to embed malware, making it necessary to employ separate, trustworthy AI checks to detect such intentional compromises.
- AI Hallucinations and Sensationalism Warning - The speakers discuss how AI‑generated reports can hallucinate sources, urging critical evaluation without resorting to sensationalist paranoia.
- AI Governance Gap Alarm - The speaker warns that about 60% of organizations lack AI governance and security policies, likening the situation to running a race without a finish line and emphasizing the chaos and potential breach costs that will ensue.
- Quantum Threats and 2026 Predictions - The hosts wrap up the segment, thank guests, and tease Jeff’s upcoming 2026 technology predictions video, highlighting recent quantum advances and AI’s role.
- Year in Review: Shiny Threat Actors - A candid discussion about the erratic “shiny happy spiders” and related hacker groups—covering false retirements, repeated FUD tactics, surprise attacks on hospitals, and new players like ClickFix.
- Social Engineering Meets Supply Chain Attacks - The speakers critique the simplistic “don’t click” warning, examine sophisticated scams like ClickFix, and note the increasing prevalence and visibility of supply‑chain attacks across platforms such as GitHub and npm.
- Breach Size Claims and Risks - The speakers mock the overuse of “largest breach in history” labels while comparing massive data leaks, noting a health‑records breach, and cautioning that browser extensions create hidden supply‑chain vulnerabilities.
- Trust Failures in Cybersecurity Episodes - The speaker recaps three podcast episodes—an alliance data breach, a patch‑management lapse involving a long‑unpatched SharePoint flaw, and insecure contractor access—highlighting trust as the common weak point and urging security professionals to educate the public.
- Session as New Security Perimeter - The speakers discuss evolving threats—from ransomware and cryptomining to token hijacking—and argue that authentication sessions are becoming the next security boundary.
- Securing Delegated AI Agents - The discussion warns that entrusting credentials to AI agents creates inevitable theft and hijacking risks, emphasizing the urgent need for strong security guardrails and defenses for both businesses and individuals.
- Observability, Change, and Shadow AI Threats - The speakers discuss how continuous change requires observability to quickly surface unknown risks, while warning that shadow AI agents and crypto‑mining misuse represent emerging threats that enterprises must proactively address.
- Resilient Crypto Amid AI Threats - The speakers examine AI agents exploiting privileges, the emerging operational risks to cryptographic systems, and express optimism about upcoming defenses such as crypto‑agility, system resilience, and AI‑powered threat intelligence.
- Closing Remarks for Cybersecurity Podcast - The host wraps up the episode by thanking panelists, guests, and listeners, urging everyone to stay safe as they look toward the upcoming year.
Full Transcript
# 2025 AI Security and Incident Review **Source:** [https://www.youtube.com/watch?v=gwL3iivVkiA](https://www.youtube.com/watch?v=gwL3iivVkiA) **Duration:** 00:57:11 ## Summary - The episode reviews the past year’s cyber‑security landscape, featuring three segments on AI & data security, incident response, and broader 2025 trends with expert panelists. - Discussions highlighted the rise of AI‑powered threats, including proliferating AI agents, “shadow AI,” and the need to both defend against AI attacks and protect AI systems from manipulation. - Specific topics covered ranged from “vibe coding” and cyber‑security sensationalism to emerging quantum threats and insights from IBM’s “Cost of a Data Breach” report. - The incident‑response segment recapped notable attacks such as the Tea app breach, ClickFix exploit, and various worm campaigns, illustrating the diverse tactics cyber criminals employed over the year. ## Sections - [00:00:00](https://www.youtube.com/watch?v=gwL3iivVkiA&t=0s) **Year-in-Review Cybersecurity Podcast** - The hosts introduce a special IBM Security Intelligence episode that recaps the past year’s AI, data security, incident response, and 2025 trends with expert panelists. - [00:03:17](https://www.youtube.com/watch?v=gwL3iivVkiA&t=197s) **Building Trust in AI Security** - The speakers stress the need for observable, resilient AI systems and emphasize trust as the central focus for upcoming AI and data security efforts. - [00:07:02](https://www.youtube.com/watch?v=gwL3iivVkiA&t=422s) **Securing AI Against Agent Attacks** - The speaker warns that as AI enables end‑to‑end attack tools, we must also focus on defending the AI systems themselves—citing Palo Alto’s discovery of agent‑to‑agent session‑smuggling vulnerabilities—and asks if the industry is giving AI protection sufficient attention. - [00:10:34](https://www.youtube.com/watch?v=gwL3iivVkiA&t=634s) **Risks of Poisoned AI Code Generators** - The speakers argue that as AI takes over more coding tasks, attackers could subvert the models to embed malware, making it necessary to employ separate, trustworthy AI checks to detect such intentional compromises. - [00:14:12](https://www.youtube.com/watch?v=gwL3iivVkiA&t=852s) **AI Hallucinations and Sensationalism Warning** - The speakers discuss how AI‑generated reports can hallucinate sources, urging critical evaluation without resorting to sensationalist paranoia. - [00:17:19](https://www.youtube.com/watch?v=gwL3iivVkiA&t=1039s) **AI Governance Gap Alarm** - The speaker warns that about 60% of organizations lack AI governance and security policies, likening the situation to running a race without a finish line and emphasizing the chaos and potential breach costs that will ensue. - [00:21:18](https://www.youtube.com/watch?v=gwL3iivVkiA&t=1278s) **Quantum Threats and 2026 Predictions** - The hosts wrap up the segment, thank guests, and tease Jeff’s upcoming 2026 technology predictions video, highlighting recent quantum advances and AI’s role. - [00:24:39](https://www.youtube.com/watch?v=gwL3iivVkiA&t=1479s) **Year in Review: Shiny Threat Actors** - A candid discussion about the erratic “shiny happy spiders” and related hacker groups—covering false retirements, repeated FUD tactics, surprise attacks on hospitals, and new players like ClickFix. - [00:27:58](https://www.youtube.com/watch?v=gwL3iivVkiA&t=1678s) **Social Engineering Meets Supply Chain Attacks** - The speakers critique the simplistic “don’t click” warning, examine sophisticated scams like ClickFix, and note the increasing prevalence and visibility of supply‑chain attacks across platforms such as GitHub and npm. - [00:31:51](https://www.youtube.com/watch?v=gwL3iivVkiA&t=1911s) **Breach Size Claims and Risks** - The speakers mock the overuse of “largest breach in history” labels while comparing massive data leaks, noting a health‑records breach, and cautioning that browser extensions create hidden supply‑chain vulnerabilities. - [00:35:15](https://www.youtube.com/watch?v=gwL3iivVkiA&t=2115s) **Trust Failures in Cybersecurity Episodes** - The speaker recaps three podcast episodes—an alliance data breach, a patch‑management lapse involving a long‑unpatched SharePoint flaw, and insecure contractor access—highlighting trust as the common weak point and urging security professionals to educate the public. - [00:38:52](https://www.youtube.com/watch?v=gwL3iivVkiA&t=2332s) **Session as New Security Perimeter** - The speakers discuss evolving threats—from ransomware and cryptomining to token hijacking—and argue that authentication sessions are becoming the next security boundary. - [00:42:17](https://www.youtube.com/watch?v=gwL3iivVkiA&t=2537s) **Securing Delegated AI Agents** - The discussion warns that entrusting credentials to AI agents creates inevitable theft and hijacking risks, emphasizing the urgent need for strong security guardrails and defenses for both businesses and individuals. - [00:48:04](https://www.youtube.com/watch?v=gwL3iivVkiA&t=2884s) **Observability, Change, and Shadow AI Threats** - The speakers discuss how continuous change requires observability to quickly surface unknown risks, while warning that shadow AI agents and crypto‑mining misuse represent emerging threats that enterprises must proactively address. - [00:51:22](https://www.youtube.com/watch?v=gwL3iivVkiA&t=3082s) **Resilient Crypto Amid AI Threats** - The speakers examine AI agents exploiting privileges, the emerging operational risks to cryptographic systems, and express optimism about upcoming defenses such as crypto‑agility, system resilience, and AI‑powered threat intelligence. - [00:56:21](https://www.youtube.com/watch?v=gwL3iivVkiA&t=3381s) **Closing Remarks for Cybersecurity Podcast** - The host wraps up the episode by thanking panelists, guests, and listeners, urging everyone to stay safe as they look toward the upcoming year. ## Full Transcript
this idea of vibe coding. Yeah, it it's
it's great until it's not. The AI, the
code that it writes, you know, is
getting better. It didn't used to be
great, but it's getting better and it's
going to keep getting better. I like to
say, don't bet against AI.
>> All that and more on security
intelligence.
Hello and welcome to Security
Intelligence, IBM's weekly cyber
security podcast where we break down the
most interesting stories in the field.
Now, today we're doing something a
little bit different. Uh instead of
breaking down stories from the past
week, we are breaking down stories from
the past year with the help of some of
our favorite panelists. Of course,
they're all favorites. These six just
said yes when we asked them to join us.
And here with me today is Patrick
Austin, staff writer for IBM Think and
the news correspondent for this podcast.
Patrick, thank you for stepping in front
of the camera with me today.
>> Thanks for having me, Matt. Uh, happy to
be here.
>> Absolutely. Now, the way the show is
going to work is we have three segments
for you all today. We've got Michelle
Alvarez and Jeff Kroom on the year in AI
and data security. We've got Dave Bales
and Nick Bradley on the year in incident
response. And we've got Suja Visen and
Shridhar Mupiti with a broad survey of
some of the big lessons, trends, and
innovations of 2025. Now Patrick, you
talked to Suja and Shridhar. Can you
maybe give us a little taste of what
your conversation covered?
>> Of course. Yeah, we um Suja Shridar and
myself touched on the topic of cyber
security software and cyber security in
the past year. It's been a very uh busy
year in terms of cyber security
especially when it comes to AI powered
um cyber security incidents. We
discussed um we discussed the
proliferation of AI agents. We discussed
shadow AI um and the and the havoc it
can wreak on a corporation potentially.
And we discussed innovations and
responses that companies can take in the
in the next year when it comes to
combating uh cyber attacks.
>> Absolutely. And that kind of, you know,
really funnels into what Michelle and
Jeff and I talked about, which was some
of those gaps and edges in AI security,
right? Like we talk a lot about, you
know, how we have to protect ourselves
against the AI and and deal with shadow
AI and stuff, but we also talked a lot
in in my conversation with with Michelle
and Jeff about the need to protect AI
from people, right? Like this is a new
kind of technology that can be like
socially engineered in a way that other
things can't. So, we covered that. And
along the way, we also touched on vibe
coding, cyber security sensationalism,
quantum threats, and of course, IBM's
cost of a databach report. can't look
back on the year without touching that
one. And then with Nick and Dave, we
dove into a kind of veritable carnival
of cyber incidents. You know, we talked
about the tea app and and click fix and
the shy hallude worms, scattered laps as
hunters, all kinds of stuff. And a big
theme that emerged for this year and
next year across both of my
conversations, and I'll be interested to
hear if you felt like this theme was in
yours, too, was this idea of trust. Who
has it? Who shouldn't have it? How do
you give it out? Did that come up in any
of your conversations at all?
>> Absolutely. I think that maybe was one
of the most uh pivotal or integral
points of the conversation. Um giving
that trust to an AI agent or or system
and you know knowing that it you can
trust it to do its job properly without
any sort of um malicious um intent. Um
we talked about observability and and
resilience as well. um and how it is
super important to be able to see what
what AI is doing behind the scenes so
you know where exactly your data is
going and how it's being treated. Um it
was it's a very huge um problem and one
that I think will you know companies
will take steps to solve in the next
year.
>> We can only hope you know but I yeah so
our our watch word for this year as we
look back and look forward is going to
be trust. And so without further ado
let's see what our experts had to say.
There was no shortage of AI security
stories this year. In fact, there has
been no shortage since chat GPT burst on
the scene in 2022. The impact of AI on
security has been top of mind ever
since. But 2025 did feel like a year
where we started to reflect a bit more
on that coverage. We started to poke and
prod at the discourse and wondered, are
we doing this right? here today to talk
about the year in AI and data security.
With me, Michelle Alvarez, manager,
Exforce threat intelligence, and Jeff
Kroom, distinguished engineer, master
inventor, AI and data security. [snorts]
Folks, to kick us off, I was hoping to
open the floor to you to tell us what do
you think were the biggest stories or
most interesting trends or moments in AI
and data security this year. Michelle,
I'll go to you first. What really caught
your attention this year?
>> Absolutely. I think we have moved from
you know contemplating our attackers
using AI and attacks to verifying right
I believe in our exforce threat
intelligence index that we published in
February of 2024 we had said we had not
seen confirmed use of AI now we can
erase that right we know that they're
actively using AI so so much has been
around are attackers leveraging how are
they leveraging
um to what extent are they leveraging
right and also um how can we use AI to
defend against attacks generally
speaking regardless if AI is being used
>> absolutely we have definitely seen like
the the threat the AI threats come from
being this thing we talked about like oh
it could happen to like they're
happening now right and I think about
that uh anthropic story very recently at
the time of recording this recently
anyway where they busted that spy ring
that was using claude code to basically
run an automated campaign of of
espionage. So like we are seeing these
AI threats actually take place now.
Jeff, how about you? What kind of stood
out for you this year?
>> Yeah, definitely the same kind of thing.
AI has gone from being a more
theoretical threat to a real threat. And
in particular, we've seen in the second
half of the year, I think, the emergence
of of agents, AI agents, as a way to
amplify the attackers capabilities. It
amplifies risk for us. Now, I'm not
against agents. Agents can do some
really great stuff if they're in the
right hands and they're under the right
control with the right governance and so
forth around them. But the reality is,
and we've seen these stories breaking
lately here toward the end of the year,
that agents have not only emerged, but
they're actually able to automate and
run independently, autonomously the
entire kill chain from beginning to end.
We we've had, you know, where it was
easier for attackers to do their thing
by click on a tool and not have to
understand the details. Now we've got
where it does the whole thing beginning
to end and that is a trend I don't
expect to see abating that that's going
to continue
>> and you know I also think about how this
year doesn't didn't just show us how
attackers were using AI and how
defenders were using AI but it also
showed us how we have to protect our AI
right and I'm I'm paraphrasing IBM's
Suja Vuison with this actually but it's
something that she said in a recent
podcast which was as much as we need to
protect ourselves from AI I we need to
protect our AI from people, right? And
one of the things that jumped to me in
that uh uh kind of realm was these agent
session smuggling attacks that Palo Alto
uncovered a little while back. I don't
know if you folks remember this, but
basically they found out there was a
vulnerability in the agentto agent
communication protocol that basically
allows you to set up a malicious agent
to socially engineer a legitimate agent
without the end user even knowing. And
this kind of thing again brings to my
mind that need to protect uh our AI as
part of our systems. I'm wondering if
you have any thoughts on that. Do you
think we focus on AI protection enough?
Michelle, I'll start with you. Matt, I
think securing AI itself is often an
overlooked aspect of AI security. Uh
most of the attention of course has been
on how AI is being used in attacks uh or
how it can be used in defenses.
uh things like we've seen of course AI
fishing, deep fakes um and we're
definitely leveraging AI uh in order to
detect attacks and identify and contain
them. But uh just as important and what
we're hopefully trying to emphasize
during our discussions with clients is
how how important it is to secure AI
applications and the infrastructure and
the data flows themselves. uh because we
do anticipate this to be an attack
surface that's continuing to grow as AI
adoption increases and therefore it will
be an incentive for attackers to target
AI technology.
>> And speaking of that expanding attack
surface, uh something that I've been
thinking about a lot and I'm not the
only person is the kind of rise of vibe
coding, right? We saw that uh term
coined this year back in February uh by
OpenAI co-founder Andre Carpathy. And
it's kind of really taken everybody by
storm now, vibe coding, right? You don't
even need to know how to code. You just
tell the AI what you want and it spits
it out. But that opens up some new
vulnerabilities in our coding process,
doesn't it? And and Jeeoff, I was
wondering if I could get your take on
vibe coding from this security
perspective. Do you think it introduces
kind of new flaws in the attack surface?
How do you feel about this?
>> Oh, yeah. So first of all the AI as an
attack surface that to me is a really
important question. It's a a
presentation I've been doing at
conference events all over the country
uh for this whole year a presentation
called AI the new attack surface and
that's one of the things I drill into
and this idea of vibe coding. Yeah it
it's it's great until it's not. The AI
the code that it writes you know is
getting better. It didn't used to be
great, but it's getting better and it's
going to keep getting better. I like to
say, don't bet against AI. I keep
hearing people make predictions about,
well, AI can do this, but it can't do
that. Well, uh, most of those
predictions, we've in fact seen AI learn
to do those things. So, even though the
code's not perfect now, you know what?
Nobody, no humans are writing perfect
code either. So, that can't be our
standard. However, to your to your
question though, as we get less and less
involved in the writing of code, then it
means if the AI were to be subverted in
some way to inject malware into our
code, when are we going to recognize it?
You know, who is going to know that
that's happened? So, that means if we're
going to have an AI creating our code,
we're going to need other AI looking
over that code. And we got to make sure
that that neither of them have been
poisoned or under the the control of of
another system. Um it's not just the
hallucinations that we have to care
about. It's some of the intentional
things. So if if I wanted to be uh
really sneaky, uh I would just
infiltrate the AI model you're using to
write your code, poison it, and then
have it do my injects for me. And then
um then you good luck trying to find
that. So that's uh that's going to be uh
certainly another thing. Uh but
we'll we'll adjust as we always have.
>> Yeah. I think that, you know, don't bet
against AI is a really good point,
right? Because I think about when again
Chad GPT, you know, came out 2022, it
was really impressive, but people kept
saying, oh, but you know, look at the
videos it makes are bad, right? And oh,
you can tell it's look at how far we've
come now, right? So you apply the same
thing to the cyber attacks and it's like
yeah obviously you don't want to bet
against the AI but sometimes I wonder if
we bet maybe too much on the AI because
another thing that's happened this year
is we've seen an increasing kind of
skepticism among some cyber security
professionals about how we cover stories
particularly when it comes to AI and the
thing that comes to mind for me there is
I don't know if you folks remember but
back in the spring MIT put out this
paper about how 80% of ransomware
attacks are are involve AI somehow. And
then uh in the fall, some cyber security
researchers including Marcus Hutchkins
and Kevin Bowmont started to poke some
holes in that methodology. They dug in
and they realized that the paper was I
don't want to use too strong a term, but
not really. It wasn't as good as they
claimed. And and that paper got pulled
down. And this is a paper that was
widely cited after it came out. So I'm
wondering if we have a sort of
sensationalism problem in the way that
we cover AI and security. Jeeoff, I'd
like to get your take on that. Do you
think we're dealing with some
sensationalism here?
>> So, we've got AI that hallucinates and
now we've got security people that are
hallucinating and making up stuff as
well. So, now maybe we can't just blame
the AI for these kinds of problems. But,
uh, yeah, I mean, there's there's always
been an issue with this. Um, there's a a
a balance where we want to make people
aware of what the threats are. But if
you go too far down the FUD road, the
fear, uncertainty, and doubt, and play
that angle up too much, well, then
people start to tune out because it's
like the the little boy that cried wolf.
First of all, nobody wants to hear that
there's a wolf, but okay, if there is,
we need to tell them. that if we're if
we exaggerate these claims and we've
been hearing about apocalyptic claims,
you know, for as long as I've been doing
security that decades now. Um, and you
know, so there the one thing that all
the the these exaggerated claims have in
common is that, you know, they're
they're not well researched. They're
probably well intended. I'm assuming
most people have good intentions,
although I'm sure some don't. But we've
got to in the era of AI, the most
important skill is critical thinking.
Whether you're using AI to generate that
information and I just read a story this
morning a colleague sent to me a major
consulting firm doing a contract and
they had used AI apparently to do some
of the the build the the case in the
report that they were giving to their
client and it turned out the AI was
hallucinating sources. So yeah, we we
can't just trust everything that comes
out just the same way that we haven't
been able to just trust everything that
comes out of the internet. So we've got
to have I I know this will be a shock.
Not everything on the internet is true.
So ju just remember that. So everything
that does come out, we need to put that
through a critical thinking filter and
say, okay, is it possible that something
was reduced by a,000%. No,
mathematically that doesn't work. So,
we're going to have to look at that
again and uh and make sure that we keep
our heads fully engaged and um you know,
look, we need to know about the
warnings, but we don't need to
exaggerate them. In fact, I would argue
we don't need to exaggerate any of this
stuff because it's bad enough if you
just tell the accurate the accurate
story.
>> Michelle, I'm wondering if you have any
thoughts on this kind of sensationalism
issue that's been maybe plaguing us for,
as Jeff says, quite some time. It's not
unique to AI, but what what have you
seen this year? Yeah, absolutely. To add
on what Jeff said, basically we want to
make sure that we are cautious and
vigilant but not paranoid. Sometimes
that's easier said than done. And often
times as security practitioners, we also
have a responsibility to not just say,
"Hey, there's a thing uh but also say
there's not a thing." And that's really
going to depend on an organization's
cyber threat landscape. It's not one
shoe fits all, right? It's going to
depend on your industry, where you
operate geographically speaking, and so
if you see something in the news and
that might fit your profile, maybe you
should raise your uh red flag, right? Um
but also partnering with a threat
intelligence partner that can help you
sort of make that decision and
prioritize what are really the threats
out there. And sometimes we just don't
know because all the facts are not on
the table when something is first
reported.
And it's not paranoia if everyone really
is out to get you. [laughter]
>> What about if every AI is out to get
you? Um, no. Let's I feel like uh no, we
can't end this segment without a a look
at the cost of a databach report because
this is, you know, IBM's annual report,
dare I say a landmark report comes out,
everybody looks at it and they should.
It's got tons of great information about
on what data breaches are like for
organizations right now. So to round out
the segment and and luckily enough I
have two people who probably know this
report better than anybody else. I was
hoping I could get your takes on, you
know, this year in data breaches. What
do you think were the biggest kind of
takeaways from this report this year?
And and what should organizations kind
of keep in mind as we head into 2026?
And um Jeeoff, I'll start with you. What
do you think?
>> So I think there were a lot of
interesting things that we could we
could pull from this. Um this is the
first report where we really started
seeing some AI related information come
out and you know it it wasn't great. You
know we're finding you know on the order
of 60% of organizations have no AI
governance and security policy in place.
Okay that's not a good thing if this is
going to be something that's going to be
core to the business and we don't yet
haven't we have not yet defined how it's
supposed to operate. What are the
boundaries? What are the ways that we
ensure that it's operating within those
boundaries? Well, then we're just asking
for for a mess. Um, that's I think in
terms of because I like to run. I've got
a race tomorrow. I I hope I know where
the finish line for that race is. But if
they, you know, just uh fire the gun and
say, "Everybody run until we tell you to
to stop." I'm not running that race. So,
this is what it's like if your
organization has no defined this is what
success looks like for us. And without
governance policies, without securities
policies in place, you don't know what
it looks like. You're running and you
have no idea what it looks like in the
end. And and by the way, I I I want
Michelle to answer and then I I want to
come back and say one thing about what I
I actually built a time machine, went
into the future and was able to read the
future cost of a data breach report and
I want to tell you what I found in that.
>> I can't wait to hear about your trip
from the future. Michelle, give us your
take. What do you what what did you pull
away from the report this year?
>> I can't top that. And Jeeoff, you've got
so many things going on between your
race tomorrow and jumping into the
future, but I do have to second Matt
your point about this being a landmark
report. And I think one of the things
that makes this a landmark report is
that every year we can count on it
telling us what are the top things that
increase the cost of a breach and what
are the top things that reduce the
costs. Uh and great news guys globally
average speaking uh the cost of the data
breach went down that's the first time
in 5 years so what are organizations
doing right and that's always what we
hear from sysos and the seauite what are
other organizations getting right we
want to do the same thing well now we
have and have had now over a decade this
report that tells us that and what I I
think are the big tech takeaway is the
um increased use of AI and automation to
detect breaches faster to detect them
and to contain them quicker which then
equates to reduction in detection costs.
Now, of course, regionally speaking, um
we have some areas that we're going to
find higher cost because of uh
regulatory costs like in the US um which
is understandable, but overall on
average costs have been reducing um and
that's some really great news to share
this year.
>> Now, future man Jeff, tell us what you
found.
>> Yes. Yes. So, I I built my time machine
and I jumped into the future and I
actually found a copy of the cost of a
databach report and and what it said was
now that we have quantum computers,
people have been breaking our crypto and
reading all of our secrets and the the
data breaches are occurring because
quantum computers can read the stuff
that we didn't we didn't uh we didn't
make quantum safe. And so, it's a
cautionary tale. Now, the only the the
thing I wish I had done is I forgot to
look at a calendar while it was in the
future. So, I can't tell you exactly
which year the report was from, and it
didn't have a date on it. So, I don't
know exactly what year that's going to
happen, but I'm just going to say that
it's the dye has already been cast. One
day, we're going to have a cost of a
data breach report where one of the the
contributions to data breach will be the
cracking of crypto. Um, and that's
something we really haven't had to worry
about in the past. So, there you go.
>> That's an extremely good point. I think
that for a long time the the kind of
harvest now decrypt later threat was
just kind of a fun thought experiment,
but we have seen a lot of quantum
advances this year and we are getting
pretty close to it. Uh, that is the end
of our segment. That's all the time we
have for today. I want to thank you Jeff
and Michelle both for being here, not
just for this segment, but for all the
expertise you've shared on our pad
podcast over the past few months. and I
really hope I see you both again on the
show many many times in the new year.
And uh one last thing before we go,
Jeff, I know you have a video coming out
with some uh predictions for 2026. You
want to tease that for the audience here
before we end?
>> Yeah, sure. So, this has become now an
annual uh thing for me coming out with a
video on the IBM technology channel
YouTube um where I talk about the first
of all I I make these predictions. So I
look back and see if my predictions were
correct from the previous year and this
is about the third or fourth year that
I've done this now and then go looking
forward into the future. And so I've
actually touched on some of those here.
Um AI is going to be a huge part as it
was last time uh in terms of the
predictions that I think will be coming.
Now since I have the time machine, you
know, I can actually go into those and
see what they are and and I know they're
all going to be 100% accurate. So that
there's that. Oh, I love it. I can't
wait to see it. Uh, but again, thank you
both folks for for being here today. My
>> pleasure. Thank you.
>> What was the biggest cyber security
incident of 2025?
Not exactly an easy question to answer
considering that any given week is bound
to give you at least one candidate. I
mean, off the top of my head, we had the
JLR attack. We had the the Tapp data
leak. We had uh scattered lapsis hunters
doing all kinds of things. And so it's
really difficult to narrow it down, but
also I'm not the cyber security expert.
I'm just a humble podcast host. And
today I do have two experts with me,
familiar faces to those who have been
watching the show. We've got Dave Bales
and Nick Bradley, both of X Force
Incident Command and hosts of the Not
the Situation Room podcast, which means
they are very well positioned to survey
what has happened in 2025 and tell us
what were the biggest incidents, the
biggest stories to pay attention to. So,
I'm going to open up the conversation to
you folks here and maybe we'll start
with you, Nick. When you look back at
everything you folks have covered this
year or seen happen, what sticks out to
you? What are the stories that you feel
like defined 2025? That is not so easy
for 2025. You know, usually when we have
a given year, there's some some given
nasty event that happened that everyone
will remember for all time. This one
doesn't really have one specific one,
but I guess if I were to if I were to
just reach into my grab bag and pull out
the ones that are going to stick with me
for this year is Clickfix, supply chain
attacks, and then everything AI.
>> That makes a lot of sense. Yeah, I was
going to say those those three the those
are the ones that that come up
constantly through just about every open
source intelligence review weekly. One
of those is going to be in there
somewhere. And then and then if we want
to talk about, you know, not so much
security events, but things that really
affected us, and that's the the too big
to fail that proved us wrong between AWS
and Cloudflare. So, uh there you have
it.
>> Absolutely. Dave, how about you? When
you look back at the year, what are the
ones that kind of pop out to you? shiny
happy spiders. [laughter]
That's what we took to calling them. Uh
the shiny lapsis uh hunters uh was
probably the the biggest story to me of
of the year because it just it went away
and it came back and it went away.
>> They threatened to go away a couple of
times and then they just came right back
anyway.
>> Promises, promises. I I think I think
the whole thing was just meant to seow
FUD, you know, fear, uncertainty, and
doubt. I'm pretty sure that's all it was
for. Don't ever listen to anything they
say.
>> Yeah. I mean, you know, I when they I
remember when they announced their kind
of retirement, I was like, "Oh, maybe
they are actually done." And and you two
both are like, "Yeah, don't hold your
breath." And sure enough, you were
correct about that. The very next day, I
feel like they were like, "We're back."
You know,
>> it pretty much was that to that level,
right? It was like they just decided to
pout because they had some of their
infrastructure taken away. Like, fine,
we quit. But we don't. Not really. But
but [laughter]
>> they were taking their ball home. You
know,
>> this time we're not going to attack
hospitals, except we did.
[laughter]
>> It's like you said, Nick, there's so
many different things that popped up and
and you mentioned ones that were also
were were were floating around in my
head, too, right? First up is is
ClickFix, right? Cuz I feel like you're
right. I saw this everywhere and we even
saw, you know, uh um evolutions of it
like I think File Fix was one and and
there might have been one called Jack
Fix or maybe I made that up. Let's let's
let's dig into ClickFix a little bit.
You know, what made that such a I don't
know an important story for you this
year aside from the fact that it just
showed up everywhere. What are your
thoughts there?
>> So, it was it was crazy successful and I
don't really get why, right? because
there's a lot of steps to it. It's not
it's not that easy. I mean, what is
ClickFix? Right? For anybody that might
be sitting there going, "Please explain
to me what is it?" Clickfix is a type of
cyber attack where you trick the user
into running the malicious commands on
their own computers for you. And in like
in some cases it'll start with like a
fake error message or security alert
that that you know convinces them to
take action and then gives them a way to
fix it and it may go all the way of
providing them the malicious script to
go to to go run on their own machine and
people do it. I [gasps] so I I'm not
sure why it was so successful. That's an
answer I can't give you. But I can tell
you that it was very successful. Yeah,
there was there was more success in that
than I thought there was going to be as
well because like Nick said, it it was
30 steps to get to one issue and you
kept looking at it going, uh, when are
these people going to fix this?
>> It relied on having people just like run
scripts on their computers, which most
of most most users don't even know what
that means, right? It's like the fact
that it worked so well that people are
just willing to open up like a part of
their computer they never go to and just
copy paste something because I don't
know they read like a YouTube video that
instructions that said they should do
that. It's a little bit crazy to me, you
know. Um I just I don't know. It feels
like maybe our our security education
doesn't work super well, but has it ever
worked super well? You know what I mean?
It
>> just when you think it's safe to go back
in the water, right, Matt? Cuz I mean
we're thinking we finally got people to
figure out don't click on stuff. Don't
click on stuff. Don't click on stuff.
But run malicious code for the bad guy.
I'm all about that. [laughter]
>> Well, you know, I mean, technically,
you're right. They're listening. They're
not clicking. They're hitting controlV.
That's a little bit different, you know.
So, [laughter]
now we have to tell them, don't click on
things, don't press buttons, maybe just
don't do anything. No, I'm kidding. Um,
but yeah, the the Click Fix was an
interesting one for that very reason. I
I wouldn't expect such a complicated
social engineering scam to work so well,
but it did. Right
now, you also mentioned supply chain
attacks, which yeah, I feel like every
single week there was some kind of crazy
supply chain attack that we saw. We had
the sales law of drift breach, which
turned into the gains site breach, etc.,
etc., etude is back now again, ripping
through npm registries or whatever. So,
let's [snorts] let's talk a little bit
about the supply chain angle here. You
know, the state of supply chain
security. I mean, did we see more
attacks of that kind this year or was it
just they were more prominent? You know,
why did why did it stick out for you
folks? Well, I think it was inevitable
because we were watching continuously
multiple supply chain attacks be it
GitHub repo, npm what you know whatever
the the storage medium is. It was the
the supply chain kept getting hit over
and over and we just saw these stories
building and building until eventually
finally shy hallude manifested itself
out of this and now we have the
self-propagating malware that's taking
advantage of the supply chain attack.
So,
I it was coming. It was It was meant to
be. We saw it on the horizon and then it
showed its ugly head,
>> and it's not going to go away. It's it's
it's going to stick around. It is going
to be probably one of the larger attack
surfaces in 26 to look out for because
it's it's relatively easy to do and it's
a it's a target base that is going to be
around. supply chain isn't going
anywhere, so it's always going to be
available to attack.
>> It was a surface that people trusted too
much, if I could say it that way, is is
they they felt like they were in a safe
environment. Everybody, you know, that's
contributing here, I guess, to the
GitHub repos or whichever repo we're
talking about or there was too much
trust
and then it came back to bite us. Yeah,
I'm glad you folks brought up this trust
angle and this kind of, you know, this
angle of this attack service not going
away because it hearkens back to
something that Dave you said. We
recorded an episode yesterday and you
were talking about, you know, I asked
our developers the new front line and
you said, "No, no, they're not the new
front line. They're the new targets of
these attacks, right? Like they're who
we're going after." Wondering if you
could expand on that angle a little bit
for our listeners who maybe didn't hear
that episode, but I also just think it's
a very important thing to say. Yeah, it
it the developers have a reputation to
uphold and when they get hit with these
uh shy hallude attacks for example, the
developers reputation is what takes a
hit and with shy hallude they're
impersonating the developer. So now
you've gotten to the point where you
don't know who to believe anymore. Uh
the developer can say hey I fixed the
code but did that actually come from the
developer? You don't know. So now
developers have to work extra hard to
rebuild their reputation, to regain that
trust that they lost because of
something that they had absolutely no
control over. Thinking back then about
all of these stories we've just
discussed, you know, your click fixes,
your supply chain attacks, your
scattered lapsis hunters, I I I know it
can be difficult, but if you could kind
of distill all of this down to some key
themes for the year, you know, and maybe
things like developers and the new
targets, I don't know, but what do you
think the key themes are that we need to
carry forward with us into 2026? Any
thoughts there? Let's start with you,
Dave.
>> Watching out for the largest breach in
history. [laughter]
Every I'm not kidding. Every single week
we got one more story that was the
largest breach in history. And you know
just some examples the Chinese
surveillance network was a 631 gigabyte
database theft um labeled the largest
breach in history. And then we had one
that was 2.9 billion records. That was
the largest breach in history. And then
we had one oddly enough that was like
1.6 six uh terabyte or billion records.
That was labeled the largest breach in
history, but obviously 2.9 is larger
than 1.6. So, we can't figure out what
the largest breach is actually going to
be, but it's going to come back.
>> Don't forget Alliance Health, Dave. That
one was almost 1.5 million social
security numbers. So,
>> and that's a big number.
>> Yeah, you don't have to put a B behind
social security numbers for that one to
be important. Yeah, you you put an M
behind that one and it's still a big
number.
>> The other thing I want to throw out is
uh that we didn't mention is browser
extensions because browser extensions
were a bit of the soup dour this year as
well. We saw those come up a lot and uh
I almost forgot to mention it, but that
still builds on the same issue we just
saw with the supply chain uh
compromises, right? Because it builds on
expected trust. people enabling browser
extensions for uh to help them with
whatever they're working on and don't
realize that you have just granted
permission to something that you really
don't know exactly what it does and that
one ends up stealing credentials,
stealing data, stealing more PII, and
the circle goes round and round.
>> Yeah, you want to get that YouTube video
downloaded, you want to pull the sound
from it, you install the browser
extension, and all of a sudden you're on
a botnet. It it hearkens back to
something we were talking about again
yesterday, which is this this we use all
these things as proxies for trust
sometimes, right? Where we're like, "Oh,
I downloaded this off the, you know, the
the extension store. It must be fine."
You don't know that. You know what I
mean? Just because it's there. I was
reading today about, you know, this this
this ring of hackers who they would
upload a perfectly legitimate uh browser
extension. It gets it passes the checks.
It gets in there and then afterwards
they sneak some bad code into there and
it goes undetected. You download that.
They're in your browser now. and and you
trusted it because you thought it was
legitimate. So, I feel like that's
another important lesson too from the
year is like you you have to do your
personal due diligence, right? Like you
you can't just assume you can trust
these things no matter where they're
coming from. You know, Matt, there's a
word that keeps coming up here and we
keep saying it and that word is trust.
It's so much implied trust. And before
this started, I decided I was going to
go jump on our YouTube channel and look
at our podcasts and see which were the
most popular ones of the year, right?
And the very first one that popped up as
the most viewed was our episode called
Who Spilled the Tea? And that one's
about the tea app that was used to uh to
for rating dates. I won't get into the
details. Rating, that's with a t rating
dates. And that one was breached. And so
everything that people were were pouring
into that app was disclosed. Too much
trust again. Because in most cases,
that's something most people probably
wouldn't share with anybody. But here
they were just sharing their heart out
about how this date went last night,
whether it was good or whether it was
bad. And now that's out for the world to
read. And let's see, the other ones were
the next one is that's private or is it?
And that one was about the alliance data
breach. Again, uh you trusted them to
handle your data and they didn't handle
it so well. Uh let's see other episodes
really quick because I know time is
probably short. Uh we had an episode
called space the final frontier and that
one was about patch management failures
and that had all to do with the
SharePoint year old vulnerability that
blew up. They chained a couple
vulnerabilities together. That one got
ugly and come to find out that
vulnerability had been disclosed
multiple of them for well over a year
and just not patched. And then lastly
was out of the top four was digital
escorts. And that one is not what you
think. And that was allowing contractors
allowing contractors in other countries,
less secure and untrusted countries to
access US infrastructure because again
we're trusting them.
>> That's going to be the running theme
through 26 is trust. And and think about
this. We're talking about breaches that
happen to companies that have security
measures in place. Think about what this
is doing to the public at large. they
don't have the same training that we do.
So, I think it's up to us as security
professionals to get as much information
as we can out there doing things like
this show uh and and letting people
know, hey, you think that you can trust
your computer, but you really can't. You
really need to stay on your toes and
follow all of this advice that's being
given to you through news outlets,
through podcasts, through uh readings
and and papers and things. I I worry
about the trust uh that the public gives
their computers more so than I do uh the
security companies or you know just
companies in general who have security
posture. It's it's going to get worse
before it gets better.
>> Absolutely. And I think that's an
incredible way to kind of wrap up this
segment. I want to thank you guys both
for being here, not just today, but for
appearing on the show so many times and
I hope we have you back a ton of times
in the new year because you guys are
great fun. And this is my personal pitch
to all of our viewers. Please go watch
not the situation room. Give them a
follow if you haven't seen it yet. If
you like what we do here, you'll love
what they do. They have even less of a
filter than we have.
>> True enough. Thanks, Matt.
>> Thanks, Matt. Appreciate it.
>> Hello, Suja Visen and Shar Mupi. Thank
you for joining me today. I'd love to
talk about the year that was um you know
2025 and the string of high-profile
cyber security incidents that we've seen
that have been having some longl lasting
effects on various companies um and just
the world in general. Um Shar I would
love to start with you um and just get
your take on on the on the past year um
in terms of cyber security and you know
how you think it went for you know in
general and maybe even specifically. I
think for me um this was a year of
silent AI sprawl right not just for u
the attackers but also for the
defenders. If you look at it from one
perspective, we had some really big
incidents like the salt typhoon which
hit um not just uh you know the
government aspect but also the private
sector like telos right on the other
side um if you look at their defenders
and folks like us u the proliferation of
AI has absolutely exponentially
increased from wipe coding to be able to
use AI for productivity. So that was
really uh heartening to see but at the
same time worries me that some of these
innovations are um ahead of security
policies.
>> Yeah sure I can see I can see that
perspective for sure. Um Suja how about
you?
>> I mean see there are two sides of the
coin with V coding comes V hacking right
and then free trials are becoming zero
day vulnerability these days. What we
have seen, we have seen anything from a
big ransomware like Land Rover Jaguar
that happened to uh a Dutch uh employee
in a windmill going and installing
crypto mining stuff as part of it and
then doing things. So we have seen both
ends of it. All kinds of crazy things
that we could have never imagined. I
think that's what technology always
does. Always throws a curveball at us
and then keep us on our toes. It's a I
mean that's a pretty smart way to start
uh mining some crypto just go go
straight to the source. [laughter] The
the other thing Patrick and Suja right
deviating slightly from the AI story I
know we've seen this this is the
identity as the next parameter I'm
beginning to see session as a new
parameter right we're beginning to see
things like you know sales loft and
salesforce situation or Gemini or even
um the echolaks right where just
establishing an authentication and
session is no longer sufficient right?
Um attackers are stealing these tokens
or credentials and reusing them in the
ways they're not supposed to be used and
hence causing some devastating results.
Right. So we'll see that more and more
in the coming years. But to me, you
know, I wanted to talk about the session
as a new perimeter right now.
>> Sure. Yeah. Um I mean I would love for
you to expand on that shar if you if you
can. Um you know what you mean by the
session being the new sort of perimeter
that we need to establish in terms of
security. I think what I mean by this
whole uh emphasis on session hijacking,
right? If you look at the sales loft um
uh attack um the user has authenticated
and established a regular or a
legitimate session and in in general we
have been used for static security where
you authenticate once you establish
security and establish a session and you
assume that things are okay and then you
start conducting business. But that may
not be okay moving forward because
attackers are now stealing those session
credentials, right? And be able to
replay that um and and and and and
use them in ways that they're not
designed for. Hence, we start we need to
start thinking about more dynamic
security than static security. Right.
Sean.
>> Sure. I'm thinking of of um you know
stuff like zero zero trust protocols to
make sure that you are who you say you
are and things like that.
>> Exactly. Right. SuJa, you know, I'm I'm
wondering if you feel the same way. Do
we need to get more granular and more
sort of um maybe more frequent on
checking up on our session activity and
see who's doing what um you know, all
the time or or do you think we should
maybe keep it the way it is? I maybe
think you're on the former side.
[laughter]
>> It's definitely need changing, right?
Whether how we with AI browsers, how we
browse are changing, how we conduct
businesses are changing from the
consumer perspective as well as
enterprise perspective because I might
log in and give access to my agent to go
conduct business. What if somebody
poisons that agents and then able to go
do it because people are doing it all
the time like find me the best deal
right so holiday scams are going in. So
people can steal the credentials.
Credential stealing is not something
new. But what Shr is talking about is in
today's world where you are delegating
it to something else, an AI, an agent.
When you're delegating, then
somebody can steal it. When somebody
steals it, it again cyber security is
not a question of if, it's a question of
when. When it happens, do we have our
defenses in place to protect our
business as well as as individuals? It
seems shar you mentioned earlier the AI
the year of AI for all essentially um
and Suja you were mentioning AI browsers
and agents um it seems like we are sort
of maybe getting a little ahead of our
skis in terms of giving these AI agents
a lot of um power permissions without
sort of establishing serious or or um
you know hard guidelines for them and
we've seen that taken advantage of in
the form of AI agent hacking and and you
know agent hijacking essentially. Um do
you think that there is hope for the AI
agent to sort of become more secure more
helpful in response to these attacks
against them?
>> We don't have a choice. I don't think
there's hope. We need to make sure that
these are secures because a year if you
ask me like 5 years or 5 years back that
will autonomous car be a reality I would
have said no no way they can co coexist
with humans. Now we are taking way more
instead of like Uber and then driving
around. So it needs security needs to
come into play. Technology is always
going to be running fast and we need to
make sure proper guard rails are
replaced and we have done that with data
right when social media and everything
came in the regulation always comes
later to catch up. the security comes
later. It's a constant learning. So
that's why I talked about the question
of when it happens, are we able to make
sure that we are secure? Are we able to
reduce the blast radius? When it
happens, are we able to observe and know
what happened so that we can prevent it
from happening future? Because without
making mistakes, you cannot make
innovation. So there is always going to
be risk associated with it. But are we
able to catch it fast enough before it
becomes too late? I think Suja captured
it really well. Right. So I mean I mean
my big takeaway Patrick for this year is
our controls are predominantly designed
for a world where tools and identities
change very slowly
right um but that is changing that is no
longer true right and that sets us that
that requires obviously new innovation
and all that but that's the biggest
observation for this year. I would love
to talk about the external pressures
that companies might be facing when it
comes to cyber security and and you know
cyber security preparedness um with a
lot of you know government and global
regulation changes when it comes to
cyber security. How do you think that is
affecting companies who want to sort of
stay at the forefront um of the cyber
security landscape but also you know
maintain their maintain their ability to
to uh you know be properly regulated and
and adhere to government regulations or
global global cyber security policy. I
think if you look at the you know the
whole geopolitical situation right you
know like you said you know there are
big attacks that are not only impacting
the government entities but also the
private sector we've seen that with you
know some of the large telos with the
salt typhoon right um so that basically
means that you know the blast surface is
not just limited to state but limited is
extending to private sector as well
right so that kind of creeps into you
talked about policy And so what what
that means is that you know some of the
policies that are impacting will will
impact the private and public
partnerships right if if there are
policies that are helping the cyber
security that's awesome but if they're
not then that's going to slow down that
doesn't mean that we give up on security
right that basically means that we just
need to think about how do we now learn
these things as a norm and build
architecture ures and systems that are
resilient for such changes right so
that's the way I would think about using
these external pressures and external
volatility as a norm and then design
system to accommodate for that
>> I think see with these government
regulations and everything things are
always going to be changing right it's a
seasonal it'll keep it will keep going
up and down so for us the bigger thing
is how do we get creative in these ways
what I see with the security industry is
security and observability coming really
really together because the CISO and CIO
and IT departments coming together to
combat [clears throat] this because when
you are deploying agents when you are
deploying AI in your in your uh
enterprise you need to be thinking about
hey what is happening that is one thing
when how did it happen what kind of
process were done to get to us to this
So when these things are coming together
then the budget becomes bigger. It's not
just about the CISO budget now the CIO
budget comes together the automation
budget comes together to think about how
do we protect at the same time make
people more productive right so bringing
those are all the things we will get
very creative when these things happen
but the mission doesn't go away because
we need when a sec cyber security breach
happens it's not just about oh something
somebody somebody's data got lost or
something bad happened it's about
reputation of a company people you lose
trust.
>> I like I like what you said Suja over
there right the observability because to
me you know one part that I was trying
to say is that okay change is is norm
right and try to figure out how to not
have a stable policy and endless hiring
forever and react to that. The other
part that you're saying is which is
really important is you know
observability budgets could increase
from one perspective. The other thing is
also what we don't know what we we don't
know what we don't know right so having
that observability helps in identifying
what we don't know very quickly and be
able to react right with precision and
accuracy right so that I I like what you
said
>> you know in the next year we're going to
see a lot of new emerging threats um
targeting companies you know public and
private sector um businesses and even
consumers um suja I'd love to know what
you think these threats might be um and
how you and whether or not you think
that we are currently uh prepared to
deal with them.
>> I think one of the thing that I would
say is about shadow agents and shadow AI
because we are all using it dayto-day in
every every uh new employee that comes
in from college. They're all AI
literate. They are used to using these
agents. So they're going to be using it.
How does enterprises prepare themselves
so that we don't inadvertently cause
harm to ourselves? like this guy who was
mining crypto mining using windmill. So
that is that is one thing. The other
part that I think about is static
encryption is dead. We with quantum
computing becoming a reality very soon
are enterprises prepared to be agile
when it comes to their crypto posture.
Shr.
>> Yeah. No, I think I think I just want to
double click on the agent part SUJA a
little bit more right because I think
you know the shadow AI and shadow agents
are already kind of here right this year
right but I think if the impact of that
right the fact that an agent may
inadverently move data without the
knowledge may cause in a data loss which
we didn't expect right imagine an agent
which is happily doing some work and and
helping and optimize a um uh workflow,
right? And as a part of that, it needs
to move data from cloud to on-prem to
different SAS applications and and
behind the scenes, the data is being
moved, manipulated in ways that we
didn't plan for. So, as a result, you
know, you may see data exposure that we
didn't plan for, right? That's what
worries me. The other part, um just
double clicking on that agent is the
identity piece, right?
uh who has access to what and what they
have done with it. Simple sentence, but
it's hard to go and prove the
accountability, right? Because agents
like you've seen with both Echolaks as
well as with Gemini, it was a simple
situation in Echolaks where the chat was
used to go and provide some cryptic
instructions and behind the scene, the
agent did what it's supposed to do. It
went and used a rag pattern to go get
all the HR data and then encoded that
into ways that you cannot even stop with
a gateway and exfiltrated that. Right?
So that's happening because the agent
was trying to do the right thing by
obtaining privileges and expanding in
what it can do to go get the data in a
rightful way. Right? So those
accountability issues will be front and
center. Um and then completely agree
with Suja and the static and Christian
right it's like it's not about crypto is
becoming so brittle that it is an
operational and a strategic risk right
it's just not a theoretical one that hey
quantum is going to come and take my job
away right it's it is it is more of an
operational thing that we have to worry
about some
>> I would love to u touch on the
innovations and sort of responses to
these adversarial attacks coming in the
next year. You know, you talk about
shadow AI and managing these these
agents, you know, operating maliciously
behind the scenes. Um, what are you most
hopeful for in the in the coming year in
terms of improving our defenses and and
maintaining that uh sort of proper cyber
security posture? Uh, Suja, I can start
with you.
>> We talked about crypto agility that is
going to be the new resilience. We need
to make sure that the systems are
resilient for the for the new world of
computing and AI. So I would say
resilience resilience are are the system
resilient to take these hits and then
still operate as good as they should be.
So that is the innovation that I'm
really excited about. I would also say
that um this AI powered threat
intelligence because human beings get
tired looking at so many things now. How
do we use this tool to make it much more
uh like it's like bad the agents need
you cannot fight agents with humans. You
have to fight agents with agents. So we
need to make sure the defense agents are
ready to change based on what the attack
surface is coming what the attack is
coming through. I I mean listen I've
been I've been in cyber security for
more than couple of decades right now
right and always worried about insider
insider attacks right I think what I'm
most excited about is the innovation
that is going to help protect the AI
right being able to treat every agent
like like a like we treat humans treat
them as first class citizens right treat
them as first class citizens so that you
identify them you provide the right
level of delegation
ensure that we have the know's point
fine grain authorization and
so that you can drive accountability
right especially as agents are drifting
and agents have autonomous behavior
driving that accountability I'm excited
for that the other piece I'm also
excited is um innovations in
observability right observability has
been not a new topic not a new topic but
agents are autonomous. Agents also
drift. So there will have to be new
types of observability that will come up
in in the next year or so that will
provide the observability for the Asian
behavior. not just understanding you
know the shadow IT and the shadow sorry
shadow agents and the shadow AI but also
what they have done with it with the
appropriate agent behavior so that we
can then figure out how do I minimize
that data exposure how do I drive
accountability how do I answer my simple
question of who has access to what and
what they've done with it um and that
will help even the crypto agility where
where Suja was alluding to
notion of making it's not about the next
new crypto cipher for we've seen that
for time in memorial right but it's
about being able to change that
dynamically right whether it's the
regulation is saying it whether it's
operational requires all that is
something which is really exciting for
2026
>> cool yeah it sounds exciting I'm excited
to have an AI agent uh co-orker that I
won't see at the uh at the office party
>> but I want to add one thing Patrick
where Shr was talking about a treat
agents as first class citizen right but
what I would say is that treat them like
humans the bigger problem is look
everybody talks about the threats of AI
what agents can do the bigger thing is
humans because how do we prevent humans
from harming these AI and agents because
that's what happens in most of these
cases they're just doing their job the
AI and agents they are telling they're
doing what we tell them to do but if how
do you prevent the bad actors from
manipulating them and then going do
that. That's where the cyber defense
come into picture big time. So the thing
that we need to be working towards is
how do we prevent humans from harming AI
and agents around good good stuff. It's
not just it's about
agent and human cohabitants.
>> You just ask the agent to get their
boss's permission. That's all. That's
that's my that's my contribution. Shar
Suja, thank you so much for your time
and for talking about the um cyber
security for for the upcoming year. Um
thank you for for your time.
>> Thank you.
>> Thank you.
>> Okay, so [snorts] that's all the time we
have for today. I want to thank all of
our panelists, Nick, Dave, Jeff,
Michelle, Suja, and Shridhar. And thank
you, Patrick, for for stepping in front
of the camera. And you know, I hope
we'll see some more of you on the show
next year. You what what do you think?
Maybe.
>> Yeah, why not? That sounds great. Thanks
for having me. I had a real blast.
>> Wonderful. I'm glad we did, too. And I
want to thank the viewers and the
listeners for sticking with us over the
past few months, for giving us a chance
to start a new cyber security podcast.
And I hope to see you all in the new
year. Let's go ahead and stay safe out
there.