1. Podcasts
  2. CyberWire Daily
  3. Keeping cyber threats on ice.
CyberWire Daily

6 Nov 2026 · 28m · 13 chapters

Episode artwork

Keeping cyber threats on ice.

In short

Cybersecurity news roundup plus an interview on how AI capability pacing won’t stop attackers.

Guests

Greg Notch, CTO of Expel (cybersecurity; focuses on offensive/defensive dynamics of AI in security).

Key claims (Notch)

Frontier AI labs slowing model development is unrealistic and may be driven by economic/regulatory positioning. Open-weight models can be “unfettered” via safety-removal techniques (“model ablation”), enabling local, unmonitored reverse engineering and exploit creation. Even if frontier progress paused, defenders would still face years of catching up. Defenders should plan for massively parallel attacks that overwhelm defenses, and be cautious deploying AI in defenses due to token-cost/DoS-like risks.

Notable examples

EU accuses TikTok of harmful “rabbit holes”; AI impersonation scheme allegedly stole ~$129M; EU/US warnings about hacking industrial refrigeration controllers; OpenAI GPT-6 Cyber model results; D-Link unpatched switch discovery-protocol flaws.

Written by AI. May contain mistakes.

Chapters

0:12

Challenges of Rapid Scaling in Security

Discussion on the complexities of managing security as companies grow rapidly.

“When you run security for a company that's scaling fast, the stakes just keep climbing.”
1:24

Cyber Threats to Refrigeration Systems

Overview of ongoing cyber threats targeting industrial refrigeration systems.

“Transcription by CastingWords switch vulnerabilities.”
2:55

Legislative Implications for Healthcare Cybersecurity

Exploration of potential legislative changes in healthcare cybersecurity.

“It is great, as always, to have you with us.”
5:30

Data Breaches and Security Challenges

Report on recent data breaches affecting various companies.

“The company attributes the intrusion to attackers using AI-driven techniques and an on-path attack.”
6:46

New Developments in AI Cybersecurity

Announcement of OpenAI's new security model and its capabilities.

“It has not disclosed how many customers were impacted.”
8:03

Regulatory Scrutiny of Social Media Platforms

Discussion on TikTok and other platforms facing regulatory challenges.

“The flaws involve authentication, access controls, and information handling in the D-Link discovery protocol.”
9:51

AI-Powered Impersonation Scheme Charges

Details about federal charges regarding an AI impersonation scheme.

“If violations are confirmed, TikTok could face fines reaching 6 % of its annual global revenue.”
12:04

Introduction to Greg Notch

Introduction to guest Greg Notch and upcoming discussion points.

“include hospitals, financial institutions, and other businesses.”
13:30

Interview with Greg Notch on AI Regulation

Insights from Greg Notch on the implications of regulating AI development.

14:00

Understanding AI Models and Their Implications

Explore the differences between open-weight and frontier AI models and their potential dangers.

“in terms of them being in pole position to be the frontier models.”
Show all 13 chapters
18:05

The Risks of Unrestricted AI Models

Learn about the dangers of unrestricted AI models and their implications for cybersecurity.

“Well, there's literally nothing to stop an ablated model running in your basement from doing that.”
22:37

Future of AI in Cybersecurity

Discuss the evolving landscape of AI in cybersecurity and the balance between attackers and defenders.

“So you've got the capabilities favors the attacker for some period of time.”
25:58

Meta's Change in Content Rights Policy

Understand Meta's controversial plan to use content from business customers and the backlash it received.

“And finally, Meta has hit the brakes on a controversial plan to expand its rights to use content from business customers' websites.”
A man walking through an airport, asking his phone a question

Ask about this episode

Get the gist, explore an idea or ask about something said.

Claude

On CyberWire Daily, what are the key points of this episode?

Searching your listening

Transcript

Automatic transcript. May contain errors.

0:01Greg Notch: You're listening to the Cyber Wire Network, powered by N2K.

0:12When you run security for a company that's scaling fast, the stakes just keep climbing. More compliance frameworks, more vendors, more risk, and a board that wants to see it all in one place. And all the information you need, it's scattered everywhere. Meanwhile, you're making this quarter's decisions based on last quarter's data. Vanta offers a better way. Vanta's Agentic Trust platform connects your compliance program with over 400 integrations and continuous monitoring. It automates evidence collection, helps identify the risks that deserve your attention, and organizes audits around your auditor's requirements.

0:51And the Vanta agent works around the clock with the full context of your program. Vanta reports a 526 % return on investment over three years, paying for itself in just three months. It's trusted by more than 16 ,000 companies, including Snowflake, Atlassian, and Ramp. Spend less time chasing compliance data and more time managing risk. Learn more at vanta.com slash cyber. That's V-A-N-T-A dot com slash cyber.

1:55Transcription by CastingWords switch vulnerabilities. The EU accuses TikTok of leading users down harmful rabbit holes, while tech giants challenge California's child safety law. Federal prosecutors charge three suspects in a$129 million AI-powered impersonation scheme. Our guest is Greg Notch, CTO of Expel, sharing thoughts on how pacing AI won't stop the bad actors. And Meta's AI feeding frenzy gets put on a diet.

2:34It's Friday, November 6th, 2026. I'm Dave Bittner, and this is your CyberWire Intel Briefing.

2:54Thanks for joining us here today. It is great, as always, to have you with us. Five U.S. government agencies are warning of ongoing cyber attacks targeting industrial refrigeration and cold chain monitoring systems worldwide. CISA, the FBI, and the Departments of Health and Human Services, Agriculture, and Transportation say attackers are targeting temperature controllers used in grocery stores, cold storage facilities, and refrigerated transportation. Attackers may exploit authentication bypass vulnerabilities, code injection flaws, and weak security configurations to remotely manipulate refrigeration settings.

3:35They could force equipment into continuous heating or defrost cycles, potentially spoiling food and medicine, damaging equipment, creating fire hazards, and putting consumers at risk. Attackers could also prevent operators from seeing accurate temperature readings. The agencies urge organizations to remove refrigeration controllers from the public internet, segment operational technology networks, strengthen authentication, patch vulnerable devices, and monitor for suspicious activity. They also recommend establishing safe temperature limits and alerts. The warning highlights how cyberattacks against relatively simple industrial controllers can have serious physical consequences.

4:22With Democrats poised to control both chambers of Congress in January, healthcare cybersecurity, AI safety, and health data privacy could receive renewed attention. Experts expect bipartisan interest in strengthening security standards and helping rural hospitals defend against cyberattacks, but disagreements over funding and regulatory costs could complicate legislation. The Trump administration's deregulatory agenda presents another obstacle. A sweeping overhaul of the HIPAA security rule proposed under President Biden remains uncertain, with final action delayed until at least July 2027. Lawmakers may also scrutinize AI use in healthcare and sensitive health information collected by apps and chatbots operating outside HIPAA protections.

5:15Industry experts warn that healthcare organizations shouldn't wait for Washington to act. Cyberattacks continue threatening patient safety, clinical operations, and sensitive data, regardless of who's running Congress. Blockchain intelligence firm Chainalysis says attackers exposed usernames, password hashes, and live authentication tokens for all customers, expanding an earlier breach disclosure involving customer identity information. The company attributes the intrusion to attackers using AI-driven techniques and an on-path attack. Chainalysis has taken several products and its login portal offline while strengthening security.

5:59Customers will be required to reset passwords and rotate API keys. The company says it found no evidence that attackers accessed investigative data or manipulated its blockchain attribution records. DigiCert has disclosed a data breach exposing business customers' contact information and, in some cases, identity documents. The incident began when an internal archiving system was inadvertently exposed to the Internet. Attackers exploited a self-registration feature lacking multi-factor authentication to access archived records between June 27th and July 2nd. DigiCert discovered the data exposure in October and has since strengthened security controls.

6:45The company says its certificate issuance systems were unaffected. It has not disclosed how many customers were impacted. OpenAI has unveiled GPT-6 Cyber, a new security model rated critical for cybersecurity capabilities under the company's preparedness framework. The model significantly outperformed its predecessor in exploitation benchmarks while demonstrating improved adherence to authorized testing boundaries. In Linux kernel research, it reproduced more than 1 ,700 vulnerabilities and developed working privilege escalation exploits for nearly 1 ,000. It also identified 81 high-severity Firefox vulnerabilities, with Mozilla already rolling out patches.

7:33OpenAI says it has reported 304 critical issues to maintainers. New safeguards monitor authorized activity and restrict assistance with attacks on critical infrastructure. Dragos helped evaluate protections for industrial systems. Access is initially limited to approved customers participating in OpenAI's Daybreak Red program with broader availability planned for authorized security researchers. D-Link is warning of multiple unpatched vulnerabilities affecting its DGS-3130 series managed switches. The flaws involve authentication, access controls, and information handling in the D-Link discovery protocol.

8:19Attackers with network access could potentially steal sensitive information, modify configurations, change credentials, or disrupt services. D-Link hasn't assigned public CVE identifiers and is investigating whether other products are affected. Until firmware updates are available, administrators should disable the discovery protocol, restrict management access, remove unnecessary SNMP community strings, and monitor for suspicious configuration changes or unexpected resets. The European Commission is accusing TikTok of violating the Digital Services Act by allowing its recommendation algorithm to draw users into potentially harmful content rabbit holes.

9:06Regulators say the platform repeatedly serves videos about sadness, self-harm, and body image, potentially damaging the mental health of young and vulnerable users. Investigators analyzed roughly 20 ,000 videos and concluded that TikTok's existing safeguards are ineffective. They want recommendations based more on users' explicit preferences rather than engagement signals like viewing time. The case marks an unprecedented challenge to the engagement-driven business model of a major social media platform. TikTok disputes the findings, calling them fundamentally inaccurate. it. The conclusions remain preliminary, and the company can challenge the evidence.

9:51If violations are confirmed, TikTok could face fines reaching 6 % of its annual global revenue. European officials say they're also examining similar recommendation practices on other platforms. Elsewhere, Meta, TikTok, and Google are suing California to block a new law designed to protect children 15 and younger from addictive social media features. Assembly Bill 1709, scheduled to take effect January 1st, prohibits platform designs the state considers psychologically exploitative. The companies filed separate federal lawsuits arguing the law violates First Amendment protections and conflicts with Section 230, which limits platforms' liability for user-generated content.

10:38They also contend that parents, rather than the government, should determine appropriate online safeguards for children. California Attorney General Rob Bonta says the state will defend the legislation. Governor Gavin Newsom signed the measure in September, citing concerns about technology's impact on children. Google warns the law could undermine existing tools designed to deliver age-appropriate content. Federal prosecutors have charged three people with using AI-generated impersonations of corporate executives and attorneys to steal approximately$129 million. Jacob Sol, Sharon Hazan, and Guy Sheetrit are accused of targeting thousands of companies worldwide over more than 18 months.

11:27According to the indictment, the defendants contacted accounting departments posing as CEOs requesting confidential financial transactions. They then used AI to impersonate executives during phone or video calls before directing employees to supposed attorneys, also impersonated by the conspirators. Victims were instructed to wire money to international bank accounts, with proceeds converted into cryptocurrency to conceal their origins. The defendants faced charges including wire fraud, aggravated identity theft, and money laundering. The alleged victims include hospitals, financial institutions, and other businesses.

12:18Coming up after the break, my conversation with Greg Notch from Expel. He shares thoughts on how pacing AI won't stop the bad actors. And Meta's AI feeding frenzy gets put on a diet. Stick around.

12:43Mimecast still protects your email, and that is not changing. But they haven't stood still. They've built around the person behind the inbox, across Slack, Teams, the browser, and the AI agents acting under their login. 8 % of employees drive 80 % of the risk, and now that includes their agents. Mimecast Apex is one platform that shows you where your risk really sits and protects the work wherever it happens. The Mimecast you trust with email now covering far more. Mimecast secures humans, data, and AI. Start at Mimecast.com.

13:29Greg Notch is CTO of Expel We sat down to discuss his thoughts on how pacing AI won't stop the bad actors So we've seen from some of the leaders of some of these frontier AI labs that perhaps we should slow the pace of the capability development of these models What's your reaction to that proposal? I don't think it's realistic. I also, in a more cynical point of view, think it's effectively a stab at regulatory capture for them in terms of them being in pole position to be the frontier models. And I also think it's an economic reaction to the costs to them for keeping pace is extremely high. And so all of those things benefit them to slow down.

14:19But unfortunately, the world of AI and frontier models are not going to slow down. Well, can you contrast this versus the open-weight AI models? For folks who might not be tuned into that distinction, how do they differ from the proprietary systems that the frontier labs are creating? I guess the best way to think about it is, or at least the way I think about it, is the open-weight models basically lag the frontier models by something like six months. So if you think about what OpenAI's best models were six months ago, those are readily available in OpenWeight, either for you to download or run on your own hardware or find a GPU hosting provider to host for you.

15:01There's this notion of model ablation, which has to do with safety mechanisms. Can you unpack that for us? Yeah, sure. So the super terrifying version of it is all of the models that are built, both open weight and by Frontier Labs, have safety controls built into them. You're not allowed to have it build a better version of smallpox for you. You're not allowed to have it build cyber weapons for you. And the way that it does this is they put the safeties on after the fact. It doesn't actually happen during model build. They're added later. And so there's a technique, it's similar to distillation where you copy a model, but where you make a copy of the model, but you remove all of the safety controls.

15:47And so then what you're left with, if you're in China, you end up with a version of DeepSeek that's willing to opine on Tiananmen Square. Also, it's willing to have discussions with you about how to build a nuclear weapon. And those things are scary because you can download an open-weight model and run it in a place where it's not monitored in any way. And so this basically unfetters you. And in a cyber context, that means you can reverse engineer software, you can build malware, you can build offensive cyber capabilities. And those are all things where both the distributed open-weight and frontier models attempt to stop that.

16:30Well, I know you've suggested that it wouldn't take a huge investment these days to operate an unrestricted AI model locally. I think that price is plummeting over time. Right now, if you want a frontier like Opus class to use the Anthropic version, so a GLM-53, for something around$200 ,000, you could run that in your house. As long as you have a free dryer plug. I mean, if you want to run a smaller model, so some of the quantized models, you can build a completely reasonable rig for about$20 ,000. And I think those prices are just going to keep plummeting, especially as GPUs are cycled out of big data centers.

17:17There's effectively no moat around that technology and the price is going to drop. That's the thing where it should give everyone pause because there's no rules about what you run on your hardware. where no one's watching. And so all of the safety guards can be easily reversed. What's a hypothetical cyber attack where an unrestricted AI model could change what the attacker could accomplish? A good example is the reverse, there's a new repo of reverse engineering software that's making their rounds now called RIA. And it's quite excellent. But if you run it and you attempt to say, take apart something and remove its cryptography or you took a look at the firmware for a medical device and it'll be like, I'm not going to do that.

18:01Like your Opus models will refuse. Well, there's literally nothing to stop an ablated model running in your basement from doing that. So the scary version of it is you're able to accomplish reverse engineering. You're able to build exploits sort of away from prying eyes and with no restrictions. and everything that you do will be novel when you weaponize it. So I found a zero day in a firewall or I found a zero day in a medical device or any of those things. Like those are all, I can weaponize the entire attack chain using AI. And so that is, you know, we haven't seen a ton of it hit the press, but it is certainly well within the realm of technical possibility.

18:50Do you suppose we're going to see that that gap between the frontier AI models and the open-weight alternatives is going to be shrinking? I can't tell the answer to that one, to be honest, because the frontier models are, they've embarked upon something called recursive self-improvement. So now the models are training the models and making them better. So I don't know enough to tell you about the pace of where the frontier models are and where they're going to go because I think predicting that is very difficult, even for the people who do it all day. What I can say is that even if we stopped the pace of progression right now, we couldn't make any model any better than Opus 5.5 or Fable or the ChatCPT model.

19:37Even just that capability, we'll spend years catching up to what we can do with those capabilities. And the open-weight models, like I said before, are maybe six months behind that. So we're kind of already past the point of where you can do amazing things with GLM and there's nothing stopping you from doing that. Suppose you're a CISO and you're looking at this emerging threat. What sort of things would you recommend that they be focusing on these days? I think there's a lot of ink spilled on the speed of attacks, like the time from recon to exfiltration, for example. And while that's true, that's not really the...

20:18Fast attacks have always been possible. it's the scale. It's you might get hit not by one agent attempting an extremely fast attack. You might get hit by 3 ,000 of them at once. And no one's defenses are really ready for that, that kind of scale. So I would encourage this to think about, well, think about massively parallel attacks. And what do you do? How are you building defenses in depth to solve that kind of problem because your defenders can get overrun. And the tangential version of that is be very careful about how you're implementing AI in your defenses or at least have eyes on them because if you look at what happened to Hugging Face, they spent 100 grand in tokens over a weekend trying to defend against this to figure out what happened.

21:12And so your budget is almost a denial of service point by a parallelized attack. So I think you have to think about your program in a completely different way, which is not so much like, sure, are my foundations covered? But what happens when like the size of what comes is bigger than anything that we've had before? And some of it will be noise, but I'll have to prosecute all of it if that makes sense. Yeah. Do you suppose that we have a version of equilibrium in our future? Or is it going to be push-pull for the foreseeable future? I don't know. Look, I operate on the blue side. So maybe this is a bit of optimism on my part.

21:57I actually think in the long enough timescale, LLMs favor defenders. Because one of the things defenders has to do is process huge amounts of data and cover lots of doors and windows if you think about it that way. And, you know, being able to deploy software and agents to do that is a massively scale-improving thing. Like, I can deploy expertise that would be very expensive for me to hire in the form of inference at my gates and doors. However, that is on a long enough time scale. I think in the short time scale, I think we have, like, you know, we're going to have all of the vulnerabilities that AI can find easily being found by people who couldn't do that before.

22:37So you've got the capabilities favors the attacker for some period of time. I mean, my crystal ball is broken in terms of how long things take anymore. But let's just say for a year, 18 months, maybe less, maybe more, we're in this place of where it's seriously favoring the offensive side. And I think defensive tools take longer to catch up. So I think we will reestablish a different equilibrium and a different baseline. I think we're still figuring out how to use AI on both sides. So there will be a new baseline. Because mainly the thing that drives cybersecurity is economics. Like, how expensive is it to prosecute an attack?

23:16How expensive it is to defend it? Like, how do I force an attacker to have a very expensive path to their goal? So I think that's what will reestablish the equilibrium. But I think it might. It's going to be a fun, by fun, I mean maybe terrifying, like, you know, interim state. That's Greg Notch, CTO at Expel.

23:53And now, a word from our sponsor, SpecterOps. 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 SpecterOps helps teams secure the AI-driven identity era at specterops.io. Thank you.

25:03software-led 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.

25:46Try it out for yourself and book a demo online at meter.com slash cyberwire. That's M-E-T-E-R dot com slash cyberwire.

26:10And finally, Meta has hit the brakes on a controversial plan to expand its rights to use content from business customers' websites. The proposed commercial terms would have allowed Meta to collect material from linked domains and webpages, potentially including news articles to train its AI products. Journalists and publishers objected to having their work swept into Meta's training pipeline without explicit permission. Meta initially delayed the changes from October 30th to November 20th, then paused them indefinitely. A Norwegian court issued a temporary injunction against the terms, while British regulators announced plans to investigate.

26:54Meta had already softened some language, removing explicit references to AI training. For now, existing terms remain in place. And apparently, even in the age of artificial intelligence, permission is still a thing.

27:15And that's The Cyber Wire. For links to all of today's stories, check out our daily briefing at thecyberwire.com. Be sure to check out this weekend's Research Saturday and my conversation with Matt Poto, Associate Triage Analyst at Huntress. The research we're discussing is titled Grand Theft Auto 6 Hype Leads to Malware. That's Research Saturday. Check it out. And hello, Maria Varmazes here. On Sunday's T-minus space cyber briefing, my conversation with Vito Askins, systems engineer in intelligence and aerospace, with his new book, Everything Was Green Until Integration. That's Sunday on T-Minus.

27:53Don't miss it.

28:21Our contributing host is Maria Vermazis. Our executive producer is Jennifer Iben. Peter Kilpie is our publisher. And I'm Dave Fittner. Thanks for listening. We'll see you back here next week.

28:51Thank you.

29:21and when. Because when critical systems are at stake, compromised access can mean downtime, disrupted production, and real-world consequences. Protect what matters most with the Zage Identity Fabric platform. Learn more at zage.com. That's X-A-G-E dot com.

From the publisher

Federal agencies warn of cyberattacks on industrial refrigeration systems. Healthcare cybersecurity could get a post-midterm boost. Chainalysis exposes customer credentials, and DigiCert discloses a breach. OpenAI's latest model learns to respect boundaries. D-Link warns of unpatched switch vulnerabilities. The EU accuses TikTok of leading users down harmful rabbit holes, while tech giants challenge California's child safety law. Federal prosecutors charge three suspects in a $129 million AI-powered impersonation scheme. Our guest is Greg Notch, CTO of Expel, sharing thoughts on how pacing AI won't stop the bad actors. Meta's AI feeding frenzy gets put on a diet. 

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

Today we are joined by Greg Notch, CTO of Expel, sharing thoughts on how pacing AI won't stop the bad actors.

Selected Reading

US Government Partners Urge Hardening Industrial Refrigeration and Cold Chain Monitoring Systems (IC3)

How Democrats' Win Could Reshape Healthcare Cyber Policy (GovInfo Security)

Chainalysis says breach exposed credentials for all customers (Bleeping Computer)

DigiCert Data Breach Impacts Business Customer Information (SecurityWeek)

OpenAI Says New GPT-6 Cyber Model Hunts Flaws Without Crossing the Line (SecurityWeek)

D-Link warns of multiple zero-days in DGS-3130 Series switches (Bleeping Computer)

EU accuses TikTok of trapping users in harmful ‘rabbit holes’ (The Record)

Social media giants sue California to overturn law shielding kids 15 and under from compulsive use (The Record)

3 Charged in Scheme to Use AI to Impersonate CEOs and Attorneys to Steal $129 Million  (SecurityWeek)

Everyone Everywhere Hated Meta’s New AI Training Terms. Now, They’re On Pause (404 Media)

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

More from CyberWire Daily

All episodes