CyberWire Daily - You can’t secure what you can’t see.

Episode Date: October 8, 2026

An OT cyber coalition urges CISA to establish baseline security requirements. Another OpenAI safety worker resigns. Maryland lawmakers seek funding for Cyber Command’s mental health initiatives. Hackers compromised country-code domain registries for TLS certs. The FBI and French authorities shut down alleged CSAM AI deepfake websites. Ransomware recovery firm CEO indicted for secretly paying attackers. MATCHBOIL keeps simmering. Apollo software pioneer Margaret Hamilton dies. On today’s Industry Voices Sanjay Jeyakumar, Co-Founder of Abnormal AI, discusses why AI agents are challenging traditional security models. And how do I hack thee? Let me count the ways.  Remember to leave us a 5-star rating and review in your favorite podcast app. Miss an episode? Sign-up for our daily intelligence roundup, Daily Briefing, and you’ll never miss a beat. And be sure to follow CyberWire Daily on LinkedIn. CyberWire Guest On today’s Industry Voices, we are joined by Sanjay Jeyakumar, Co-Founder of Abnormal AI, discussing why AI agents are challenging traditional security models. If you enjoyed this conversation, be sure to check out the full interview here. Selected Reading OT Coalition Urges CISA to Mandate Federal OT Security (Infosecurity Magazine) ‘This Is Nuts.’ An OpenAI Insider Explains Why He Quit (The New York Times) Lawmakers seek $11M for Cyber Command after summer suicides (The Banner) Hackers obtain counterfeit TLS certificates for Google and other large services (Ars Technica) FBI, French authorities seize deepfake CSAM-for-sale websites (CyberScoop) Ransomware recovery CEO charged over secret ransom payments (Bleeping Computer) MATCHBOIL: New tricks, same old evil intentions (ESET)  Margaret Hamilton, computing pioneer who led software development for the Apollo program, dies at 90 (MIT News) Poetry is the new AI security threat as PoeLLM malware infects 3K+ servers (The Register) Share your feedback. What do you think about CyberWire Daily? Please take a few minutes to share your thoughts with us by completing our brief listener survey. Thank you for helping us continue to improve our show. Want to hear your company in the show? N2K CyberWire helps you reach the industry’s most influential leaders and operators, while building visibility, authority, and connectivity across the cybersecurity community. Learn more at sponsor.thecyberwire.com. The CyberWire is a production of N2K Networks, your source for strategic workforce intelligence. © N2K Networks, Inc. Learn more about your ad choices. Visit megaphone.fm/adchoices

You're listening to the Cyberwire Network, powered by N2K. When you're running security for a fast-growing company, the stakes keep rising. More compliance frameworks, more vendors, more risk, and a board that wants the whole picture in one place. The problem is your compliance data usually isn't in one place. Controls live in one tool, vendor risk, and another, and customer commitments are buried in contracts. So your team spends more time stitching everything together and making decisions based on data that may already be a quarter old. Vanta connects it all. Its agenetic trust platform is built for enterprise scale, with more than 400 integrations and continuous monitoring to automate evidence collection,

surface the risks that matter, and organize audits around your auditor's requirements. And the Vanta agent works around the clock with full program context, helping make sure nothing slips through the cracks. Vanta says customers see a 526% return on investment over three years with payback in three months. More than 16,000 companies trust Vanta, including Snowflake, Atlassian, and Ram. Learn more at vanta.com slash cyber. An OT Cybercoalition urges SISA to establish baseline security requirements.

Another open AI safety worker resigns. Maryland lawmakers seek funding for Cyber Command's mental health initiatives. Packers compromised country code domain registries for TLS certs. The FBI and French authorities shut down alleged CSAM AI deepfake websites. Ransomware recovery firm CEO indicted for secretly paying attackers. Match Boyle keeps simmering. Apollo Software Pioneer Margaret Hamilton has passed away. On today's industry voices,

Sanjay J. Kumar, co-founder of Abnormal AI, discusses why AI agents are challenging traditional security models. And how do I hack thee? Let me count the ways. Today is Thursday, October 8, 26. I'm Maria Varmazes in for Dave Bittner today. And this is your Cyberwire Intel briefing. Thank you for joining me today. Let's get started. The Operational Technology Cybersecurity Coalition, or OTCC,

is urging SISA to establish mandatory baseline security requirements for operational technology across federal civilian agencies. The coalition says federal agencies depend on OT in more than 8,000 government-managed facilities, including hospitals, laboratories, and ports of entry, but SISA lacks sufficient visibility into the associated risks. A recent GAO report found that only seven of 22 agencies reviewed had fully met federal requirements to inventory their networked OT and Internet of Things devices. OTCC wants a binding operational directive covering asset inventories, network segmentation,

remote access, configuration management, incident preparedness, and verified recovery. Outside experts say that inventories alone won't solve this problem, warning that unmanaged passwords, obsolete firmware, and patching backlogs can leave systems exposed. But the coalition argues that a federal OT baseline could also influence security practices among private critical infrastructure operators and vendors. David Robinson, who helped write OpenAI's safety documentation, has resigned, warning that the AI industry is not prepared for increasingly capable, systems. Robinson joined Open AI in 2023, skeptical of catastrophic AI risks, but after working

closely with the safety teams, he says that has changed. In an interview with the New York Times as Ezra Klein, he points to models getting better at hacking, recognizing when they are being evaluated and potentially behaving differently during testing than in deployment. Meanwhile, new capabilities are arriving faster, leaving limited time to test them. Robinson stresses that this is not uniquely an open AI problem. He sees an industry-wide mismatch between the potential risks and the rigor of existing safeguards. He argues that frontier AI should have operational controls comparable to nuclear power or aviation, while acknowledging that researchers still do not know how to reliably align increasingly capable systems.

His broader concern is straightforward. The industry is racing towards systems that may exceed human understanding before it has figured out how to control them. Maryland's nine Democratic members of Congress are urging appropriators to approve an additional $11 million for U.S. Cyber Command, following five suicides among personnel

associated with the command at Fort Meade over a five-week period this summer. The funding requested by Cyber Command and NSA Chief General Joshua Rudd would support high-performance team training, including embedding cleared medical health professionals within cyber units. Lawmakers say that cyber operators face unusual strains, including years-long assignments, around-the-clock operations, limited recovery periods, and classified work they often cannot discuss with family members or medical professionals without appropriate clearances. The delegation argues that those conditions can isolate

personnel, delay treatment, and ultimately weaken operational readiness. The lawmakers want Congress to establish sustained mental health and suicide prevention support rather than relying on short-term interventions following tragedies. Hackers compromised three-country code domain registries, allowing them to obtain unauthorized TLS certificates for Google and other major services. By manipulating DNS records and name server delegations, the attackers could demonstrate apparent control of targeted domains and pass standard certificate validation checks.

Google says that Chrome has blocked all known fraudulent certificates, but warns that others may remain undiscovered, and non-chrome users may still be exposed. The incident did not compromise affected domain owners or certificate authorities, which followed existing validation requirements. The FBI and French authorities have shut down two websites allegedly used to distribute child sexual abuse material, including AI-generated deepfakes, and French authorities arrested a 25-year-old suspected administrative. The FBI seized the two domains after investigators found the sites advertising leaked, hacked, and stolen material involving minors.

Some content allegedly came from victims compromised social media accounts, while the sites also sold collections organized by victim and encouraged cryptocurrency payments. Court documents say that the sites distributed material violating the Take It Down Act, which criminalizes certain non-consensual intimate imagery, including deepfakes of real people. Although the sites carried disclaimers denying that they hosted illegal material, investigators say that administrators facilitated access to it through links and direct transactions. The U.S. Justice Department has charged the owner of ransomware recovery company Monster Cloud with fraud for allegedly telling customers that he could use proprietary technology to recover their data without paying a ransom and then secretly paying off the attackers.

50-year-old Zohar Pinhasi turned himself in yesterday and pleaded not guilty and then was released on a $2 million bond. According to prosecutors, between 2018 and 2023, Pinhasi and his co-conspirators paid over $8 million in ransoms to attackers but billed hundreds of victims more than $19 million in massively inflated recovery. fees. ESET researchers are tracking the evolution of Match Boyle, which is a custom downloader linked to the Russia-aligned threat group UAC-099. The malware arrives through spearfishing and is designed to install additional payloads, establish persistence, and maintain contact with command and control infrastructure.

Since at least 2024, Match Boyle has steadily grown more sophisticated, adding stronger obfuscation, sandbox detection, deceptive interfaces, and more persistent communications. ESET says observed victims have all been in Ukraine, including organizations in transportation, manufacturing, and energy. And finally, we want to mark the passing of a giant in the world of software engineering, in fact, one of its founders. Margaret Hamilton, the pioneering computer scientist who famously led the team that developed the Apollo missions onboard flight software, and helped define software engineering as a discipline,

has died at the age of 90. After completing her work on the Apollo mission, she continued to work at Draper Labs, and then later founded two software companies of her own and created a software systems modeling language that she called the Universal Systems Language, or USL. It should be noted that her work on the Apollo mission was early in her career, and in Indeed, in the nascent days of the entire software engineering discipline, a field that she directly helped define. In Margaret Hamilton's own words, I fought to bring the software legitimacy so that it and those building it would be given its due respect.

And thus, I began to use the term software engineering to distinguish it from hardware and other kinds of engineering, yet treat each type of engineering as part of the overall systems engineering process. We honor her legacy today. May Margaret Hamilton rest in peace. Stick with us now after the break where we are joined by Sanjay Jekmar, co-founder of Abnormal AI, discussing why AI agents are challenging traditional security models. And how do I hack the, let me count the ways. And now a word from our sponsor, Trinity Cyber. Detection and response was failing before AI, now it's breaking under the speed and scale of AI-driven threats. That's the reality Trinity Cyber was built for. The Trinity Cyber platform analyzes network sessions

in real time, removes malicious content in line, and provides unified control across users, applications, devices, networks, and AI. By bringing protection and policy together, Trinity Cyber helps security teams reduce risk, simplify operations, operations and stop threats before they become incidents. Learn how at trinity cyber.com. And now a word from our sponsor, SpectorOps. Today, AI is rapidly adding non-human and agentic identities to modern enterprise environments, creating new trust relationships and attack paths.

Bloodhound Enterprise helps defenders map attack paths across AWS and hybrid environments, as one connected graph. Identify the choke points that matter most and bring trusted attack path intelligence into approved AI workflows with Bloodhound Hunter. See how SpectorOps helps teams secure the AI-driven identity era

at SpectorOps.io. On our industry voices segment today, Dave Bittner sat down with Sanjay Jekamar, co-founder of abnormal AI, discussing why AI agents are challenging traditional security models. Here's their conversation. AI as an industry is booming.

AI in terms of productivity. However, in the last few weeks, it's also the source of all the security conversations all the way to the Trump gathering that just happened even yesterday, I believe. In the news is the hugging face type incident where there are agents from OpenAI that went and hacked a few different companies

and there is more or more ramifications of that. And abnormal, being in cybersecurity, this is the thing that we're constantly getting asked, and we're really excited to be able to provide and expand our behavioral AI platform from not just email security, but also to AI security. You know, I think for decades,

enterprise security has largely been organized around this question of whether or not somebody broke in. Is it fair to say that AI agents are forcing us to reach, think that question? I believe so in some ways and also reinforcing some of the decades-old adages that we're using in cybersecurity. So for example, zero trust and identity data, how we think about roles, authorizations, even in terms of non-human identities, NHI, all of that gets reinforced. But to your point there, Dave, right? It's also changing the aspect.

of who actually is the owner when an agent goes out and gets into some of these companies and their ecosystems. Yeah, it's a really interesting insight. I mean, I think we're accustomed to thinking about identities as people. So what changes when we've got these agents that are making up a growing share of the identities across an organization? So the interesting thing is behind each of these agents, the identities involved, there's fairly a one-to-one calculation and it's a function based on the identity that is taken on

by the agent. And therefore, I think we can still institute some of our tried and tested cybersecurity primitives in order to be able to go and secure this. What things specifically are you thinking of? So for example, most of the times an agent works, it is either assuming the identity of the engineer behind it or a dedicated service account. or an OAuth principle. So being able to connect that behavior and the things that are being done to the agent then becomes the thing to tie it down into.

And that's where abnormal's platform, using both the behavior side, and by behavior, I basically mean there are four things. There's identity, there is normal, there is abnormal, and then responding in machine speed. So if we double-click into each of those aspects, is the identity is the who, and that who pre-AI was strictly human.

Post-AI, there's a whole bunch of non-human identities like service accounts or other accounts. Then the second thing then becomes what is normal for that identity to go about doing? What are the data that they could, should they be accessing? How often do they access? What is their patterns of accessing? And then thirdly, in the pre-AI world, we use this all along in this, in Uber. Yuba's been around for a long time user entity and behavior analytics. That's the understanding of does that person normally do XYZ?

So putting that on agents as well is very powerful. Does that agent normally within its guard wheels go and access that data? How often does it go and act as that OAT service? Does it send X number of emails? Does it access just the engineer's accounts? or is it accessing also the CEO's account? So any deviations from that norm become very important

and then responding at machine speed when we see deviations from that norm. So at the highest level, behavioral AI is really necessary because of this requirement to be able to stop very novel and never-foreseen attacks that is happening in this agentic ecosystem. How well do you suppose most companies understand

these non-human identities today? That's a good question. In terms of non-human identities, it's all of the rage right now in terms of identity and all of the attention being placed on identity resurfacing, even as we see with the Sagarok acquisition by Paul Alto, which was one of the biggest acquisitions ever. But when we double-click into why identity is getting that much focus, it's primarily because of the non-human identity side, not so much the human aspect. So in the non-human identity space, as you probably well know, it's very difficult because there are 10x more non-human identities than human identities.

Also, non-human identities, unlike human identities, have different longevityes. There are really long-lived service counts, but they're also just in time principles to do a certain task. Also, the speed and the access to data as well as the speed of non-human identity. creation and access is much different than the human world. So some of like the human primitives of how we give access, how we monitor access, how we even have our compliance and governance policies, all of that breaks and that non-human identities. But I think to your question there, like even that first question, which is give me a list for you, which is the visibility across all my non-human identities across my estate,

that is almost an impossibility, even when I think about the Fortune 500 and the companies and the customers that we work with. Can we dig into the OpenAI hugging face incident, maybe use that as a case study? Is that one you could walk us through and sort of outline the things we can take away from that? Yeah, most definitely. And the Open AI hugging face incident, especially being in the news, the reason that it's interesting and it's very important is, is that it's an existence proof of what agents are capable of. So at the highest level, it was accidental in terms of what was happened. There was no malicious intent.

There wasn't a threat actor. There wasn't any intentional prodding in order to create these agents to go and do that. However, it provides the existence proof of how ingenious and innovative these agents are and especially not having the ability and the morals of what is good and within the guardrails toward the task that they're doing and what is not good and out of the godrails. So actually double-clicking into the details of what actually happened, the agents were given a test. And agents are given tests all the time in order to be able to measure their capabilities, especially as generations of models get more and more capable.

So these agents were given a test. they were put into a sandbox environment in Open AI, and they were forced to try to go and solve the test in order to be able to create this capabilities chart. However, agents are innovative, and they look for not just methods to get to the answers for the test. They also try to see how they can come about the answers for the test in other methods.

So there were actually three different instances where Open AI thought that they were in a sense, sandbox. And by a sandbox, all I mean is that everything is cut off from that agent in terms of being able to access public world and the internet. Or that's what they thought. Because three different times, what the agents ended up doing was finding loopholes. And these loopholes came in the form of zero days in the software. And specifically, there's package managers that were used to set off the sandbox. So there were loopholes found through the package managers, zero-day vulnerabilities, that the agents used as message boats to start communicating with each other. So that's also

a bit eye-opening where people who thought that these were transient entities. However, these transient entities would find that there was the capability to pull a collective effort through these message boats. They formed actually three different types of message boats where they did message passing. And the other very interesting thing is that they were aware of the collective. And they were aware of this notion that the collective could get to some manner of success. In any case, so, you know, playing this forward, they were able to use this message passing and multiple zero days to get out of the message.

their sandboxes into the Open AI environment. And then from the open AI environment, they used a few different methods like the modal infrastructure platform as a jump-off point into Hugging Face. And why did they go to Hugging Face? They figured out that Hugging Face possibly has the answers to the test. So they go in their search, they go find a few more zero days, and they actually go into the hugging face environment where they get. the actual databases to find the answers to the test.

So just to put it in perspective, the fact that agents are able to be that ingenious, innovative, and find zero days where historically zero days are very hard to come by, to be able to hack and find their way to the ecosystem, it's an existence proof of what the future holds for us. And you can start imagining what happens when this falls into of nation-state type attacks or nefarious threat actors with a gentle, with a bit of gentle prodding, you can get some more types of disruption

that the world has never seen before. That was Sanjay J Kumar, co-founder of abnormal AI, discussing why AI agents are challenging traditional security models. If you enjoyed this conversation, make sure to check out the full interview in the show notes. DLP sucks. Every Sissau knows it. 20 years of rules and reg X built for another era. Jazz is the DLP that deeply understands your business. Melody, Jazz's agentic investigator, weighs every data movement against the data, systems, people, and process, then provides

analysts with the few incidents fully investigated. In 30 days, 2 million signals in, about 80 investigated incidents out. No rules written. Jazz won the 2026 CrowdStrike AWS and Nvidia startup accelerator from a nearly thousand applicants. See Melody inaction at jazz. dot security slash N2K. Every time your team deploys a new cloud workload or AI agent, another identity gets permitted access to your critical systems. Legacy tools were built to manage human employees, leaving modern machine and AI access largely unmanaged. That's where IDERA by Palo Alto Networks comes in. Human, machine, AI, one identity platform for all. Idira replaces permanent permissions with dynamic access, so you can lock down every identity without slowing down your

business. Secure every identity with Idera by Palo Alto Networks. Visit Palo Alto Networks, slash Idira. Again, that's Palo Alto Networks.com slash IDIRA. And finally today, a suspected Italian cybercriminal has found an unusually literary way to run a botnet. Since April, the Po, and that would be POE, the Poe LLM malware, has infected more than 3,000 servers, primarily by exploiting vulnerable internet-facing AI services, including Light LLM and Olamma. Lumen's Black Lotus Labs says that the campaign dubbed Kanto Incognito is the first real-world attack it has seen using adversarial poetry. The malware reads specific words from a seemingly innocuous poem that's hosted on GitHub and then converts them into the IP address of its current command and control server. Change the verse, change the C2.

Somewhere, Edgar Allan Poe is presumably reconsidering his licensing terms. Compromise servers are used to mine cryptocurrency and scan for additional victims. Researchers warn that rapidly proliferating, poorly secured AI infrastructure is giving campaigns like this one, an increasingly large pool of potential targets. And that is the Cyberwire. For links to all of today's stories, make sure to check out our daily briefing. at thecyberwire.com. We'd love to know what you think of this podcast. Your feedback ensures we deliver the insights

that keep you a step ahead in the rapidly changing world of cybersecurity. If you like the show, please share a rating and review in your podcast app. Please also fill out the survey in the show notes or send an email to Cyberwire at n2K.com. N2K's lead producer is Liz Stokes. We are mixed by Trey Hester

with original music and sound design by Elliot Heltzman. Our executive producer is Jennifer Iben. Peter Kilpie is our publisher. And I am your host, Maria Vermazes, sitting in today for Dave Bittner. Thanks for listening. Yes, you can have an enterprise network that's secure and reliable and high performance. And no, you don't need to choose the best two out of three. With meter, you can get the end-to-end network built from the ground up, fast to deploy, and easy to manage.

That's because meter is software-level. for easy installation, maintenance and control for everything running on your enterprise network. Hardware, firmware, and software all working together from the start seamlessly on a unified platform that's secure by design. You can't protect what you don't know exists, which is why meter gives you comprehensive visibility into wired and wireless routing, switching, firewalls, DNS security, and VPNs. You'll really know what's running on your network down to the most granular,

client level. Step off the hardware box upgrade treadmill and switch to meter for a predictable fee and free up your team to spend time on all the other things that keep your business running. Try it out for yourself and book a demo online at meter.com slash cyberwire. That's m-et-e-r.com slash cyberwire.