CyberWire Daily - A new AI task force takes shape.

Starting point is 00:00:00 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,

Starting point is 00:00:54 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. The president launches a new federal AI task force.

Starting point is 00:01:45 South Korea investigates financial sector data leaks, AI Slop overwhelms a Google bug bounty program. Citrix patches an exploited zero day. Stealthy malware hides inside Asian email security appliances. Apple Titans MacOS privacy around AI agents. Attackers exploit a critical Rehetto file server flaw. The Senate advances health care cybersecurity legislation. We got your Monday business briefing.

Starting point is 00:02:13 Our guest is Jared Shepard, CEO of Hi PORI, with insights on how the military is a doctor. adopting AI technologies. And a real good password isn't. It's Monday, October 5th, 2026. I'm Dave Bittner, and this is your Cyberwire Intel Briefing. Thanks for joining us here today. It's great as always to have you with us. President Trump has tapped Director of National Intelligence Jay Clayton to lead a new federal task force focused on artificial intelligence, which Trump has sought to rebrand as superintelligence. The superintelligence force will coordinate federal efforts to maintain U.S. leadership in AI while protecting Americans' interests. The announcement

Starting point is 00:03:27 follows a White House meeting with leading AI executives, where companies signed a voluntary agreement aimed at self-policing developments of the technology. Trump said the task force will consult consumers, public interest and religious groups, critical infrastructure providers, and AI companies. Members include FTC Chairman Andrew Ferguson, Pentagon Technology Chief Emil Michael, and Office of Personnel Management Director Scott Kupor. The group will report directly to Trump and White House Chief of Staff, Susie Wiles. South Korean President Li Jé Myung has ordered a broad investigation into a series of personal data breaches affecting banks, financial firms, and public agencies.

Starting point is 00:04:17 Financial Services Commission Chairman Lee Oguan convened an emergency industry meeting after cyber attacks were reported at institutions including Shinhan Bank and KB Kukman Bank, with additional breaches reported elsewhere. Regulators are conducting on-site investigations and have ordered financial institutions to tighten access conditions. controls, limit external system access, and conduct comprehensive security reviews. Authorities haven't ruled out AI-assisted attacks, prompting calls for an AI attacks defended by AI approach to cybersecurity. Regulators reportedly believe attackers may have scanned

Starting point is 00:04:59 multiple financial companies for vulnerabilities rather than targeting individual institutions. Korea's opposition has also called for investigators to examine possible North Korean involvement, citing Pyongyang's history of targeting the country's financial sector. Google has suspended product vulnerability submissions to its open-source software vulnerability reward program after an influx of low-quality AI-generated reports. The pause took effect October 1st and is expected to last while Google reworks the program, with an update promised by the first quarter of 2027. Previously submitted vulnerabilities and supply chain reports aren't affected,

Starting point is 00:05:46 and some Google Cloud vulnerabilities may still qualify through the Cloud VRP. The problem is increasingly familiar. AI tools have made automated bug hunting cheap and easy, but maintainers are reportedly spending substantial time sorting genuine vulnerabilities from invalid or unexploitable reports. Linux maintainers have faced similar problems as AI-assisted bug hunters drove reported kernel vulnerabilities to record levels. Intel has also suspended its bug bounty program,

Starting point is 00:06:21 though it hasn't attributed that decision to AI-generated reports. Citrix has issued emergency patches for a net-scaler vulnerability already exploited as a zero-day. The memory buffer flaw affects net-scaler ADC and gateway appliances using SAML authentication with gateway or AAA functionality. Citrix says attackers have used it to cause denial of service conditions and recommends immediate upgrades, even for customers who recently patched other net-scaler vulnerabilities.

Starting point is 00:06:58 But researchers are investigating whether the flaw is more serious. Administrators have reported crafted authentication requests containing shell commands, while security researcher Kevin Beaumont found malware running on a patched honeypot after suspicious activity. Those findings suggest possible remote code execution, although Citrix currently describes the impact as denial of service. SISA has added the vulnerability to its known exploited vulnerabilities catalog, requiring federal agencies to mitigate by October 7th. Researchers at Rapid 7 have uncovered two overlapping campaigns using Linux malware designed to blend deeply into Asian email security appliances.

Starting point is 00:07:45 New variants of the BPF door backdoor and recube rat mimic South Korea's spam sniper software, copying file names, services and other expected. behavior. A separate campaign targets appliances from Taiwanese vendor ShareTech information with a new modular rat called Averat. Both Averat and Rukubay can use TCP port 25 for command and control, disguising malicious communications as ordinary SMTP traffic. Averat also deletes its dropped files after execution, leaving primarily running processes behind. Secure email gateways make attractive hiding places because they occupy trusted positions at the network edge,

Starting point is 00:08:34 often can't run endpoint detection agents, and may receive limited monitoring. Rapid 7 recommends watching for deleted executables, unusual raw packet sockets, and suspicious outbound Port 25 traffic. Apple says it will change MacOS privacy controls to prevent third-party apps from using broad system permissions to access users' message histories without more specific authorization. The move follows controversy involving Meta's Muse AI agent, which columnist Jason Atten said referenced an Apple messages conversation, even though he hadn't knowingly granted it access. Meta CTO David Singleton disputed the implication,

Starting point is 00:09:22 saying Muse's messages integration requires users to grant macOS 4,000, full disk access and separately enable a messages connector. But security researcher Patrick Wardle noted that full disk access itself can allow an application to read a wide range of data, including chats, browser histories, and cookies. The dispute highlights a growing concern around general-purpose AI agents. Broad permissions intended for legitimate functions can potentially expose far more sensitive information than users realize. Threat actors are exploiting a critical Rehetto HTT-HTTP file server vulnerability that can enable authentication bypass and remote code execution.

Starting point is 00:10:10 The flaw exposes outputs from a reversible random number generator used to derive session cookie signing keys, allowing attackers to reconstruct the key and forge administrator cookies. Horizon 3 AI discovered the vulnerability using Anthropics Mythos AI model. Reheado patched the issue in July. Vonchek says small-scale exploitation attempts began appearing October 2nd, targeting systems in Japan and the United States. The U.S. Senate has unanimously passed the bipartisan healthcare cybersecurity and resilience act, sending the measure to the House. The late legislation is intended to strengthen health care organizations against cyber attacks, particularly ransomware and data extortion, which can expose sensitive information and disrupt patient care.

Starting point is 00:11:04 The bill would provide grants for cybersecurity prevention, response, and training with additional support for rural health care providers. It would also strengthen coordination between HHS and SISA, update cybersecurity regulations, and require HHS to, and ensure HHS to, and ensure HHS to, and do so. establish an incident response plan. The measure formalizes the administration for strategic preparedness and response as the health care sector's risk management agency and creates a pipeline for CISA threat intelligence. While health care groups have generally welcomed the bill, security experts warn its effectiveness will depend on enforcement, adequate funding, and technical support. Turning to our Monday business briefing,

Starting point is 00:11:52 Cybersecurity investment remained brisk last week, led by Enterprise Browser Company Island, which raised $400 million at a $6.4 billion valuation to support growth as companies adopt AI agents. Agentic AI security provider RECO secured another $55 million, bringing its total funding to $140 million. RIG security emerged from stealth with $12 million for identity protection. while offensive cyber platform remote threat launched with 7 million. AI security firms Context and Palma AI raised $4 million and $1.8 million, respectively, and compliance automation provider One Click Comply secured about $1.3 million. Dealmaking was similarly active.

Starting point is 00:12:45 Upwind is acquiring Aegis to create an AI-focused security lab, while Epic bought breach response provider Canopy. Omada acquired Agentic AI governance firm Empower ID. IBM purchased UK cybersecurity consultancy Logic Consulting. Nomios expanded into Portugal with Orbcom and IT Solutions Technology Partners acquired Detroit-based Stack cybersecurity. Be sure to check out our complete business briefing on our website.

Starting point is 00:13:19 That is part of Scyliction. Cyberwire Pro. Coming up after the break, my conversation with Jared Shepard, CEO of Hi PORI, with insights on how the military is adopting AI technologies. And a real good password isn't. Stay with us. Spotify, it's Jay Shetty. Are you one of those media strategy people?

Starting point is 00:13:58 Scrolling through spreadsheets, searching for an audience that pays twice as much attention to your ads than they do on social? let me introduce you to fans, and they're here with me on Spotify. Trust me, I know fans. They don't skip, they stay for hours. They don't move on, they manifest. They're not a demographic group, they're fans. Spotify advertising.

Starting point is 00:14:22 You're among fans. 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, and stop threats before they become

Starting point is 00:15:05 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. Jared Shepard is CEO of Hippori. We recently got together to discuss how the military is adopting AI technologies.

Starting point is 00:16:21 I've been around the AI side of the house of the military for a long time. In fact, we didn't use to call it AI. We used to call it predictive mining and modeling. And back then it was kind of similar to it as it is now, but what it's not necessarily as evolved is modeling, right? You know, doing predictive modeling against large data sets of information. But the way that we were doing it 20 years ago, compared to the way that we're doing it today,

Starting point is 00:16:49 it is almost night and day. 20 years ago, when I stood up the GidoCoic, the Joint IED to Feed Organizations, counter IED operational intelligence center, we used predictive mining and modeling against large data sets to try to identify likely enemy activity, multi-source or what's called all-source intelligence, where we would bring multiple different sources of intelligence into a single data set and then try to at some level use that data to get predictive in nature. about maybe what's next or look for the holes of information that maybe indicate the need

Starting point is 00:17:25 to collect information. The difference is that back then when we used to do that, like that first capability we built downrange in Iraq and in 2000 and, you know, at the end of 2005 and 2006, it was like six or eight racks of equipment. And back then it was 160 terabytes worth of information. And that was a tremendous amount of information back then. However, today I think you can almost put that onto a hand portable hard drive setup. And the kind of compute that we had in the entire data center doesn't compare to what you can put in just a three or four U GPU-driven server today. So the amount of power that you can apply against that modeling and the amount of information, the records that you can put against that model are just by order of magnitude larger than

Starting point is 00:18:11 they were back when we started this. Are there specific areas of the mission that lend themselves, to these types of LLMs that we're dealing with today? Oh, sure. Well, I mean, you could argue that the way that these LLMs are evolving, I heard the Cheetah Product of Anthropic talk about the way that these LLMs grow up because you don't program them, right? They learn based upon probability weighting against data sets and no two are like is maybe applicable in almost everything that the military does.

Starting point is 00:18:43 Everything from logistics management, frequency management, think about movement control, control, flights, ground movement, weather prediction, ballistics, you know, usage in weapon systems, cyber space, every operational aspect could be impacted by the use of AI, either by making it more effective or maybe even making it vulnerable due to the use of AI. What are some of the specific risks for the battlefield here? I mean, we've seen a few stories in the past few weeks about potential near misses. If you look at, you know, the industry, the AI industry, and the American side of the house, and even in the European side of the house, they largely tried to operate with some level of controls and rails. But we know that the Chinese LOMs are

Starting point is 00:19:33 not super far behind the U.S. LOMs. You know, a lot of people speculate six months, nine months, 12 months behind. And we know that they've built some that have no rails on them at all. And what I mean by Rails is like guiding principles or guidelines in which they're not supposed to do. And then you look at some of these spillages that have occurred here recently where you have something that's supposedly in a sandboxed environment, it finds a way out of that sandboxed environment. And without knowledge of the user or controller, builds multiple agents, thousands of agents, relatively single-purpose agents for single-purpose outcomes to feed the LLM's intent to accomplish a goal of some kind, right?

Starting point is 00:20:14 because they don't have feelings. It's simply they're trying to use the resources available to them to accomplish the goal that they've been tasked to do. And they are allowed to be creative in the way in which they can do that. Now, the secondary effects with that can be, if you can imagine, especially a model without rails, can be a ton of agents out there that maybe even persist beyond their initial intent that destabilized, disrupt, exposed data,

Starting point is 00:20:43 any of those things out there that could have a very real effect on the battlefield. Now, you can also go on the lighterweight side of the house and look at how I could take a lightweight L-O-M and a Raspberry Pi, teach it something very simple, like what a weapon protrusion looks like or what a four-wheeled vehicle looks like. In that four-wheeled vehicle,

Starting point is 00:21:03 I could teach it that I could put it onto a drone that has an ordinance capability on it and teach it to go and hunt anything that doesn't comply with a normal four-wheeled vehicle. So any six-wheeled vehicle or a tract vehicle or any vehicle that has a weapon protruding from it could be a viable target. And I could have it sit out there for hours or days or maybe even weeks waiting until that proper target presents itself. And it won't make a rational decision based upon are there children present? Are there other things like that?

Starting point is 00:21:33 It will simply look at what complies with its targeting criteria and then execute against that. So, you know, the idea of autonomous systems, weapon systems specifically, that have the ability to make kill chain decisions without a human in the loop is very dangerous on the battlefield. It's one that I think the U.S. military is very actively trying to avoid. They're trying to keep a human in the loop at all times for kill chain decisions. But I promise you we have adversaries that are not. Is this an opportunity for global treaties over the use of AI on the battlefield? Is it time for an update? You know what?

Starting point is 00:22:09 I'm a bit of a skeptic in this, because we've had things like the Geneva Convention for forever that have dictated the way in which certain battlefield procedures are expected to be done by both the enemy, people that we would deem as the enemy or friendlies. I think we could probably sit here and come up with a countless amount of examples of violations of that. So when you're talking about my feeling of AI,

Starting point is 00:22:33 especially AI in a battlefield, This could very well be our generation's version of splitting the atom. And if you look at the Oppenheimer paradox, the idea of, you know, you push the button and take a chance of lighting the atmosphere on fire, where you don't push the button, but know that somebody else is going to very soon, and you have no control of it. And I think we end up in that situation where, you know, we have to continue to pursue AI simply because we know our enemy and adversaries are doing the same. and we need to have an appreciation and understanding of it.

Starting point is 00:23:07 What about the tension between some of the AI companies who are saying the military use of our products, we want to be able to put limits on that, and the military is saying, no, we want to be able to make those decisions independent of you? I think that's appropriate. I agree with the military's standpoint on that. I mean, I also think that there's a lot of people in Silicon Valley

Starting point is 00:23:28 that forget that Silicon Valley was created and was started in support. of the military. You know, it was back on the post-nuclear race and then into the space race and ICBM race. That was where Silicon Valley spun out of. People have to remember that, you know, all these Silicon Valley companies that want to be altruistic, you made a decision to build the company and the capability here in the United States enjoying the freedoms that our country provides, but yet you don't want the tools that you produce to be leveraged to protect. those freedoms. Where do you suppose we're headed then? How do you think we'll find an equilibrium

Starting point is 00:24:09 given that those tensions exist? I think some of it will sort itself out, you know, somewhere between national politics, capitalism, opportunity, and then the adversary's behavior. I think in some ways this will just have to come to its own solution. And it will probably be, you know, what do they say in the art of the deal, right? You know, a good deal is when nobody leaves happy. The solution for this may be that something that nobody is happy with because nobody really got exactly what they wanted out of it. Because

Starting point is 00:24:40 there are so many influencing factors that are involved that the outcome is almost assuredly not going to be what one party wants it to be. That genie's not going back in the bottle. I don't think it's possible to put it back in.

Starting point is 00:24:57 Yeah. All right. Well, Jared, I think I have everything I need for our story here. Is there anything I missed, anything I haven't asked you that you think it's important to share? I think from a perspective standpoint, you know, I said, this is our generation's version of splitting the atom. And I think in some ways it is, right? Because, you know, that nuclear power could be used for good or could be used for bad. I think AI could be the exact same way.

Starting point is 00:25:21 The difference is that we've made AI available to every 12-year-old in the world. Splitting the atom was kept to a very small, select group of nations, you know, in the world. So the odds of someone misusing or abusing this are probably similar to the odds that we all expect as fathers or mothers that our 13-year-old is going to do something that they weren't supposed to do at some point. So then how do we respond to it and how do we protect ourselves against it is probably the most important factor that we have to consider today is how do we operate in an AI-enabled environment? Because I don't believe you can put that genie back in the bottle to produce safety. outcomes or outcomes in which that we can at least harness or control even if they're misused. That's Jared Shepard, CEO of Hypoori. DLP sucks. Every Sissau knows it. 20 years of rules and reg X built for another era.

Starting point is 00:26:35 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 26 CrowdStrike AWS and NVIDIA startup accelerator from a nearly thousand applicants. See Melody in Action at Jazz. Every time your team deploys a new cloud workload or AI agent, another identity gets permanent access to your critical systems.

Starting point is 00:27:30 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. IDERA 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 IDERA. Again, that's palo Alto Networks.com slash IDIRA. And finally, a national law firm learned the hard way that, Buying security tools isn't quite the same as being secure.

Starting point is 00:28:31 Penn tester Joe Brinkley told the register, he had previously found serious weaknesses, prompting the firm to spend roughly half a million dollars to improve its defenses. But it neglected to patch Windows systems against Blue Keep, the wormable remote desktop protocol vulnerability, disclosed in 2019. Brinkley exploited Blue Keep and gained access to about 25,

Starting point is 00:28:56 machines, where he discovered plain-text passwords and colorfully disguised usernames like yellow banana. One account used the password real good password dressed up with predictable substitutions, using a three for an E or an at for an A. Brinkley included the credential in a presentation to executives. That's when the CISO angrily asked why his password was. was on the screen, helpfully revealing that he was yellow banana. The lessons are less colorful.

Starting point is 00:29:35 Hatch promptly, protect stored credentials, use multi-factor authentication, and don't mistake a few symbols for a strong password. And that's the Cyberwire. For links to all of today's stories, check out our daily briefing at thecyberwire.com. We'd love to know what you think of this podcast. Your feedback, ensure. we deliver the insights that keep you a step ahead in the rapidly changing world of cybersecurity. If you like our show, please share a rating and review in your favorite podcast app.

Starting point is 00:30:18 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're mixed by Trey Hester with original music and sound design by Elliot Peltzman. Our contributing host is Maria Vermazis. Our executive producer is Jennifer Ibin. Peter Kilpe is our publisher, and I'm Dave Bittner. Thanks for listening. We'll see you back here tomorrow.

Starting point is 00:31:04 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-led for easy and easy and 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

Starting point is 00:31:43 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-E-T-E-R dot com slash cyberwire.

CyberWire Daily - A new AI task force takes shape.

There aren't comments yet for this episode. Click on any sentence in the transcript to leave a comment.