You're listening to the Cyber Wire Network, powered by N2K. The Five Eyes issue a rare joint warning on China. Jenny Sterley weighs in on Trump's AIEO. Researchers warn everyday notifications can become AI attack vectors. Iron Worm is a sophisticated Rust-based info stealer targeting software developers. Cisco patches a critical vulnerability in its unified communications manager platform. Anthropic maps AI-enabled cyber activity to the MITRE attack framework. Authorities dismantle an online counterfeit identity marketplace. Our guest is Jason Kikta, CTO from Automox, discussing AI vulnerabilities, real risk, and the speed problem.
And an extortion crew is forced to open a customer support ticket. It's Thursday, June 4th, 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.
The United States and its Five Eyes intelligence partners, including the United Kingdom, Canada, Australia, and New Zealand, have issued a rare joint warning that China is using LinkedIn and other professional networking platforms to recruit individuals with access to sensitive government and military information. Officials say Chinese intelligence services increasingly rely on fake profiles, fraudulent job postings, and potentially AI-enabled tactics to target military personnel, intelligence officers, and security professionals. According to the advisory, operatives often pose as recruiters, consultants, or think tank employees to attract candidates who may have access to classified information. Individuals who engage with these offers can face growing pressure to provide non-public information, often in exchange for payment disguised as compensation for research or consulting work.
The warning marks the first time all Five Eyes members have jointly addressed this threat publicly. Officials say those caught in such schemes have faced criminal investigations, job loss, and revoked security clearances. The alert comes amid ongoing concerns about Chinese espionage activities targeting Western governments and institutions. In a New York Times opinion piece, former CISA director and current RSAC CEO Jen Easterly argues that President Trump's new executive order on artificial intelligence is an important first step toward addressing the growing national security risks posed by advanced AI systems. The order creates a voluntary framework that gives the government early access to review powerful AI models before public release and establishes a federal clearinghouse to coordinate cybersecurity vulnerability discovery and remediation.
Easterly contends that AI is rapidly transforming cybersecurity by enabling the discovery of previously unknown software vulnerabilities at unprecedented scale. While these capabilities could help defenders improve software security, they could also provide adversaries with powerful offensive tools if released without safeguards. She points to recent decisions by leading AI companies to limit access to advanced cyber-focused models as evidence that the risks are being taken seriously. While supporting the executive order, Easterly argues that voluntary measures are insufficient. and calls on Congress to establish durable requirements for testing, security, and deployment controls on the most advanced AI systems. CISA acting director Nick Anderson said Wednesday that the agency expects to issue guidance by the end of the week outlining the steps federal agencies must take to implement the president's artificial intelligence executive order.
Researchers at SafeBreach disclosed a now-patched vulnerability in Google's Gemini voice assistant that could have allowed attackers to manipulate the AI through malicious instructions hidden in messaging notifications from apps like WhatsApp, Slack, and SMS. The attack, dubbed fake context alignment, used indirect prompt injection techniques that Gemini could process without the user's awareness. Researchers demonstrated potential impacts including controlling smart home devices, launching Zoom calls, sending deceptive messages, and poisoning the assistant's long-term memory. Google patched the flaw in November of last year with improved content classifiers. SafeBreach says the research highlights the growing risk of prompt injection attacks as AI assistants become more deeply integrated into everyday devices and services.
Researchers at JFrog Security report a malicious NPM package uncovered IronWorm, a sophisticated Rust-based info-stealer that targets software developers, particularly those in crypto and Web3 ecosystems. The malware spreads through compromised NPM packages that execute a hidden binary during installation. Then steals credentials, cloud tokens, AI API keys, cryptocurrency wallet data, and GitHub access. Using stolen credentials, it modifies repositories, inserts malicious code into software packages, and republishes infected versions to NPM, creating a self-propagating supply chain attack. The malware includes advanced features such as per-string encryption, Tor-based command and control, and an eBPF rootkit that hides processes and network activity on Linux systems. Researchers link the campaign to compromised GitHub accounts across multiple organizations and identify dozens of backdated malicious commits designed to appear legitimate.
Despite its sophistication, operational mistakes including embedded debugging artifacts and the hard-coded cryptocurrency wallet recovery phrase provided valuable insight into the malware's capabilities and operator. Cisco has patched a critical vulnerability in its unified communications manager platform that could allow remote attackers to gain root privileges. The flaw stems from a server-side request forgery vulnerability that can be exploited through crafted HTTP requests, enabling attackers to write files to the underlying operating system and potentially escalate privileges. Cisco is aware of public proof-of-concept exploit code but has not observed active exploitation.
The issue only affects systems with the WebDialer service enabled, which is disabled by default. Administrators are urged to apply the latest security updates or disable WebDialer until patching can be completed. Anthropic has published a new analysis mapping AI-enabled cyberactivity to the MITRE attack framework, based on 832 malicious accounts banned from Claude between March 2025 and March of this year. The study found attackers used AI across all 14 attack tactics and 482 techniques, with medium and high-risk actors increasing from 33% to 56% over the year.
Most threat actors used AI for malware development, defense evasion, and data collection, but the highest risk actors leveraged AI for post-compromise activities, such as credential dumping, lateral movement, and web shell deployment. Researchers argue that traditional measures of attacker sophistication are becoming less useful as AI lowers technical barriers. Instead, the greatest risk comes from agentic systems that autonomously chain together attack stages. Anthropic says existing threat frameworks do not adequately capture these AI-driven behaviors. and is working with industry partners to develop new methods for tracking and defending against increasingly autonomous cyber operations.
French and Spanish authorities have dismantled an online marketplace that allegedly supplied counterfeit identity documents to migrant smuggling networks across Europe. The operation led to the arrest of a suspect in Alicante, Spain, and the seizure of equipment used to produce fake documents, along with roughly 800 forged European identity cards and permits. Investigators say the platform sold both physical and digital fraudulent documents that helped migrants evade border controls and obtain residents' rights. Europol noted that document fraud remains a key enabler of migrant smuggling and other criminal activities. supporting illicit networks operating throughout the European Union and the Shenzhen area.
Elsewhere, a coordinated operation involving law enforcement agencies and major technology companies disrupted more than 1.4 million accounts and infrastructure linked to scam networks operating across Southeast Asia. The effort, known as Disruption Week, targeted fraud compounds in Cambodia, Laos, and Burma that allegedly used trafficked workers to conduct online scams. Authorities arrested 63 individuals, dismantled servers and network infrastructure, disrupted social media, Microsoft, and Starlink accounts, and froze more than $3.8 million in cryptocurrency tied to criminal operations. Officials said the operation demonstrates the impact of collaboration between governments and the private sector in combating large-scale online fraud. Coming up after the break, my conversation with Jason Kikta, CTO from Automox. We're discussing AI vulnerabilities, real risk, and the speed problem. And an extortion crew is forced to open a customer support ticket. Stay with us. Hit pause on whatever you're listening to and hit play on your next adventure. Get double points on every qualified stay. Life's a trip. Make the most of it at Best Western. Visit bestwestern.com for complete terms and conditions. This episode is brought to you by TELUS.
Scammers. They're about as annoying as this. Tell them to buzz off with Telus. Get built-in scam protection included on new mobility plans and help protect yourself from pesky scam texts, calls, and emails at no additional charge. Visit telus.com slash scam protection to learn more. Conditions and exclusions apply. Hold music. Not what you want to hear when a pipe just burst and your basement is flooding. That's why Intact Insurance answers 2 million claim calls per year within 20 seconds. So instead of hearing this lovely track, you'll hear this in seconds. Hello, Intact Insurance.
Jason Kikta is CTO at Automox. And in today's sponsored Industry Voices segment, we discuss AI vulnerabilities, real risk, and the speed problem. Well, how do you assess this moment that we find ourselves in right now? We've got AI coming at us, it seems like, from every direction. And my sense is that a lot of folks are just trying to keep their head above water. What are the people that you're talking to, what are they saying that their main challenges are these days?
I think the biggest thing on their minds at the moment that's really been a consistent theme, Dave, is they have challenges orienting on the problem. And what I mean by that is, you know, they're trying to decide, is this a patching speed problem? Is this a shift from patching to mitigations? Is this a detection issue? Is this a governance issue? And the answer is a little bit of all of the above. You know, it's the same thing when I talk to people about...
you know, positive uses of AI about how it changes maybe the software development life cycle. My point is always that it's not a new cycle. It's just rebalancing of what's existing. And in much the same way, from a security perspective, the speed of certain things changes, the focus, but the, you know, overall fundamentals are still consistent. It's just where you're putting your effort and your balance.
Well, you mentioned overall fundamentals. And to what degree do you find people being able to successfully separate the reality from all of this hype? That's the one that comes up a lot as well is, you know, and my point to them is mythos, no mythos, the trend is directionally true. And it doesn't much matter whether it's, you know, chat GPT 5.5 with a really good harness or some self-trained model or, you know, somebody somehow gets access to mythos ahead of time. You know, mythos is the catalyst. It's not the threat, right? And so.
These models are accelerating discovery. They're not inventing new classes of exploitation, but the gap that really matters is between the bug being found and the posture being updated. That's what's gotten a lot faster, and that's the thing we need to focus on. So what are you recommending? How are you suggesting that people come at this? You know, I am a first principles kind of person. And so I always go back to we need to think about, you know, these frontier models and how they change the way that we balance risk. So for me, it's it starts with frontier paced governance. That's the response. You know, it's not a tool. It's a mindset shift, right? You want to align the governance of your processes with the execution around this rebalanced risk appetite, right? You're going to take a little more risk in some areas to.
potential outages the business to adjusting your network's posture than you would have in the past. And that's because, you know, that risk of disruption is far cheaper than the risk of exploitation. So that's where you need to get focused on how quickly can you identify mitigations, develop your own implementation of it and apply it. How quickly does the patch come out? How quickly can you apply that patch? You know, the thing that I'm hearing from some of my peers is that their mindset is we need to be prepared for 10x the patching. Well, that's a tall order, right? And that's an estimation.
You know, that 10 XP, that is not small. That is a very significant lift and change of how we've done business for decades. Well, you hear folks talking about having to operate at machine speed, and I'm putting air quotes around that. Again, you know, trying to separate the hype from reality. To what degree is that the reality today? I think it's the reality to a degree. You know, I teach a class down at a grad school class at Johns Hopkins Seiss, and I always like to tell my students, you have to separate what computers are good at from what humans are good at and make sure that your solution always aligns to our respective strengths and not our weaknesses, right? So humans are good at experience, judgment.
intuition right computers are good at scale speed repetition things that that humans just are naturally poor at so when you think about moving at machine speed. It's about the human applying that experience and judgment ahead of time and giving the machine, the software, the whatever, the guidance and tools it can apply at machine speed to respond to an emerging situation, right? So it's not about picking out every individual patch. It's about patching via policy and, you know, aligning your execution with that policy with no manual review, right? It's speed without breaking things. And so, you know, as an example, at AutoBox, we've put a lot of work into our blueprint.
questionnaires in our turnkey process to give people this you know get those definitive answers and that judgment and experience applied against you know notional questions and targeted questions about how they want their network to perform like you know explain this categorically rather than in the specific and then taking that and being able to turn that into an automated process that'll just run again and again and again without fail without ever slowing down or getting tired and and that's that's how you do the shift that's how you do it in a way that's reasonable and not taking some ai model hooking it up to all your systems and saying you know claude take the wheel That's not, that is a way to do it at machine speed, but I'm pretty sure even Anthropic would say, please don't do that. Right, right. Do you find that there are folks out there who find themselves kind of paralyzed by indecision? This is such a big shift that they're afraid they're going to make the wrong choice? I do, I do. And, you know, what I tell them is like, don't spend time debating about whether AI is going to accelerate attackers. Don't debate about, well, is this attack going to get faster than that? You know, focus on what you can control and what you can measure.
rather than a lot of unknowns, right? So what's your time to patch? What's your time to remediate? How are you controlling your configuration drift? Where's your governance friction, right? Can you measure your governance friction? Those are the numbers that are going to decide whether you effectively reduce your exposure or just talk about it. So focus on those things, measure them, and then drive to improve those numbers. And they're going to help you out really no matter how this unfolds. Like there is no permutation of reality where improving your time to patch or...
controlling configuration drift won't help you in the future. It even helps with IT. Like, that's the thing is that, you know, if you have patched an up-to-date machines and they're all configured the same way, well, you know, golly, that's really, really helpful when you're on an IT help desk, right? It's helpful if you're on an IT help desk and you're trying to troubleshoot something. It's really helpful when you're on the security team and you're trying to search for anomalies in the network, you know, looking for the malicious, you know, looking for the thing that doesn't fit is really hard when, you know, things are in an inconsistent state, right? So those fundamentals, that's what's going to drive you forward. Yeah.
Are there common elements that you find among the people who are seeing success here, the people who are making this transition with the least amount of drama and chaos possible? They all have a few things in common. And I think one of the biggest is they have a risk-aware, not risk-adverse mindset. So they are aware of the risks. They are aware that making any change to a functioning network is always a little bit fraught and has the potential to cause inadvertent outage or degradation of services.
There's also that risk of getting compromised, you know, having significant business impact, loss of reputation, so on. So you want to be aware and balance those two rather than try to avoid one risk or the other. Another characteristic is, you know, a heavy degree of automation. And again, I go back to moving at machine speed, whether it's through automated systems, autonomous systems that are using agentic AI. It's all really boils down the same thing of I'm going to apply, you know, my human thought process.
to a template and then tell the machine what to do. Right. And so can you automate those responses? You know, is that captured? Will it function the way you intend? Have you tested it? Right. And that's the thing is, you know, it's like I do with my security team is, you know, they'll show me some new detection that they've written up or some new tool that we've implemented. And I'll say, that's neat. Like, let's throw malware in front of it. And let's let's see what it does. You know, like throw in the thing that you're expecting it to catch and make sure it really catches it. Right. And then throw on some things that are maybe benign, but similar and see if you accidentally catch those two. And, you know, because.
You know, well, I mean, that brings me to the third point, which is nothing. is sufficient if it's not tuned, right? I don't care if it's IT software. I don't care if it's security software. You know, just implementing the thing out of the box is not at all. That's not at all going to work for you. And just no reality is it going to work right out of the box. I wish it were so. I think we all do. But, you know, our entire profession exists because, you know, it's not just the setup. It's the tuning and the maintenance. And, you know, there's another thing is that, you know, did you just set it up initially and move on? Or is it, you know, do you have a plan to go back and check on it every once in a while?
Have you done something like a solutions optimization call to with your vendor to sit down and go through and make sure that it's running the way that it ought to be? Is it aligned to best practices? Was it aligned to best practices when you brought the thing in three years ago and it's no longer aligned to best practices today? Right. Those are the critical questions that you need to ask yourself, because, you know, it's that sort of fundamental. There are those fundamental discipline things, those first principles that really carry you through. And it makes the path very, very clear. And those are the folks I see with the most success right now. That's Jason Kikta from Automox.
Hit pause on whatever you're listening to and hit play on your next adventure. Get double points on every qualified stay. Life's a trip. Make the most of it at Best Western. Visit bestwestern.com for complete terms and conditions. The impact of a workplace tragedy affects people in ways few realize. Support can be limited. Families are left to cope on their own. It can literally tear a family apart.
But support exists. If you know anyone affected by a workplace tragedy, help them mend with Threads of Life. Be the first thread in helping a family put life back together. Share threadsoflife.ca with them to learn more. This episode is brought to you by TELUS. Scammers. They're about as annoying as this. Tell them to buzz off with TELUS. Get built-in scam protection included on new mobility plans and help protect yourself from pesky scam texts, calls, and emails at no additional charge. Visit telus.com slash scam protection to learn more. Conditions and exclusions apply.
And finally, a ransomware affiliate recently learned one of cybercrime's unwritten rules the hard way. Don't target organizations in the Commonwealth of Independent States. After an affiliate tied to the RA Lord-linked Nova ransomware operation reportedly compromised the Ariel Group, an oil field services company headquartered in Uzbekistan, the victim contacted the gang to point out the mistake. Nova responded with an unusual public apology, banned the affiliate, and promised free recovery assistant and pledged not to leak any stolen data. The incident highlights a long-standing reality of the ransomware ecosystem, where many Russian-speaking groups avoid targeting organizations in Russia and neighboring CIS countries. Threat researchers say local authorities often tolerate financially motivated cybercrime, provided domestic organizations are left alone.
The episode also serves as a reminder that cybercriminals are not always criminal masterminds. Recent blunders have included hackers falling into honeypots, ransomware developers hard-coding encryption keys into malware, and coding mistakes that rendered victims' files unrecoverable, even after paying. Sometimes the villains defeat themselves, saving defenders the trouble. And that's The Cyber Wire. 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 ensures 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. 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 Keltzman. Our contributing host is Maria Vermazis. Our executive producer is Jennifer Iben.
Peter Kilby is our publisher, and I'm Dave Bittner. Thanks for listening. We'll see you back here tomorrow.