Worming its way through WeChat.

0:00You're listening to the CyberWire network, powered by N2K. 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 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.

1:03Step 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. Researcher's build a self-propagating attack against WeChat, threat actors move toward multi-agent AI frameworks, enable issues and emergency patch for a maximum severity bug under active exploitation, microtick patches multiple router OS vulnerabilities, China accesses restricted American technology through subsidiaries and overseas partners, an active fishing campaign abuses legitimate Google services to make malicious links appear trustworthy.

2:09Australians may soon be able to disable the algorithm. We've got your Monday business briefing. Our guest is Jason Lancaster, chief investigations officer at Spy Cloud, discussing how AI is affecting shifting cybercrime from traditional investigations to agentic AI at scale, and electromagnetic eavesdropping gets personal. It's Tuesday, September 8, 2026. I'm Dave Bittner and this is your CyberWire Intel briefing. Thanks for joining us here today. It is great as always to have you with us. Researchers at Palo Alto Security Company Calif say advanced AI models help them build a

3:14self-propagating attack against WeChat in little more than a week. Dubbed WeWarm, the proof of concept, exploited a zero-day vulnerability that could compromise an account when a call from an already compromised contact simply rang. No click required. The worm could then access messages, make calls, control the account, and automatically target saved contacts, potentially spreading exponentially across WeChat's more than 1.4 billion monthly users. Researchers said additional vulnerabilities could have turned account access into full device compromise across both iOS and Android. 10 cent confirmed the vulnerability and patched it after Calif's disclosure, saying it had no evidence users were affected. Calif emphasized that AI didn't create the attack autonomously. Human researchers guided the process. Still, the demonstration suggests AI is dramatically accelerating the discovery and weaponization

4:19of software vulnerabilities. Google's Threat Intelligence Group says threat actors are moving beyond simple AI coding assistance toward multi-agent frameworks capable of automating multiple stages of cyberattacks. In one incident, a financially motivated attacker used AI agents to build and launch a mass credential harvesting campaign in under six hours. The agents scanned for vulnerabilities, collected thousands of credentials, trouble shot failures, rotated IP addresses, and used compromised cloud infrastructure to evade detection. Researchers also discovered an automated framework managing more than 23,000 stolen secrets. State-backed groups are experimenting with similar technology for exploitation, reconnaissance, monitoring, fishing, malware development, and other operations. Google cautions that fully autonomous hacking isn't yet widespread, and researchers haven't

5:21observed autonomous pipelines discovering zero days and exploiting real-world networks. But increasing automation is shrinking defenders response windows by taking humans out of more of the attack loop. Enable has issued an emergency patch for a maximum security remote code execution vulnerability in its end-central management platform. The vulnerability carries a CVSS score of 10 and can be exploited without authentication. Multiple on-premises builds are vulnerable, while hosted instances have already been patched. New reporting indicates attackers are actively exploiting the flaw, creating unauthorized administrator accounts, and using end-centrals remote control capabilities to pivot into managed endpoints. Customers should install the hotfix immediately and restrict or disconnect exposed consoles until patched.

6:22Microtik has patched six router OS vulnerabilities, including two actively exploited flaws, dubbed micro-trick. CERT Poland says attackers are chaining an SSH authentication bypass with a privilege manipulation vulnerability to take full control of devices exposed to the internet. Attacks have been observed since at least September 2nd, with compromised routers sometimes containing an account named OPS. More than 120,000 micro-tick devices recently had SSH publicly accessible. Users should update router OS immediately and block SSH access from untrusted sources. A New York Times investigation found that INSPIRR, a Chinese technology giant blacklisted by Washington over its military ties, has continued accessing advanced American technology through subsidiaries and overseas partners. After INSPIRR was added to the US entity list in 2023,

7:26its Silicon Valley subsidiary began operating as IVREZ and continued exporting billions of dollars in advanced equipment. From April 2024 through February 2026, records show IVREZ shipped at least $5.6 billion in technology to Southeast Asia, including more than $3 billion in systems containing Nvidia's cutting edge blackwell chips. Some equipment went to data centers serving major Chinese technology companies, while other servers apparently traveled through Malaysia before reaching China. The investigation suggests INSPIRR's network has exploited gaps in export controls involving subsidiaries, company addresses, and remote access to overseas data centers. Federal officials have reportedly examined IVREZ, but the status of that inquiry is unclear. No before threat lab says an active fishing campaign is abusing multiple legitimate Google

8:29services to make malicious links appear trustworthy to security tools and victims alike. Attackers route targets through Google properties, including meat, search, double click, custom search, image search, tag manager, and analytics before reaching credential harvesting infrastructure. The final fishing page is personalized in real time, pulling the victim organization's logo, website imagery, and email domain information to mimic a legitimate login portal. The campaign also checks email domains and uses anti-analysis steps to filter out sandboxes and researchers. Victims are then sent down one of two paths, credential theft with stolen passwords x-filtrated through telegram or installation of screen connect for persistent remote access. No before says the campaign's strength is that nearly every intermediate step looks legitimate, reducing the usefulness of simple domain-based blocking.

9:33Australia's government plans to require social media platforms to let users over 16 turn off algorithmic recommendations and instead see content only from accounts and groups they choose to follow. The proposal called my feed my way would use a pop-up asking users to select their default feed with fines exceeding $100 million Australian for non-compliance. Separate digital duty of care rules would require social media services, games, apps, and AI chat bots to protect miners from harms including pornography, eating disorder content, misogyny, glorification of crime, bullying, and serious mental distress. Australia's e-safety commissioner would gain removal powers and require platforms to document their safeguards. The legislation is expected before Christmas, but faces political debate over censorship, enforcement thresholds, and whether opt-out mechanisms

10:37go far enough. Turning to our Monday business briefing, cyber security investment and deal-making remain active. Digital Identity Verification Company SoCure raised $156 million at a $5.2 billion valuation and acquired fraud and compliance automation firm Fravity, whose technology will be integrated into SoCure's risk OS platform. Israeli AI agent supply chain security startup Air emerged from stealth with $50 million while vulnerability remediation company Data Agent launched with $10 million. Mobile Fraud Prevention Startup Kazimi raised $2.6 million. Eight acquisitions and spin-offs were also announced. Hallowato Networks acquired Agenteg Workflow Platform Console to expand automated security operations. With SoCure spun off its sales for security business as an independent company.

11:40A line acquired security firm Pathfinder while Echo purchased technology assets from the shuttered secure container provider Minimus. Ampcus acquired security data platform SMARTER D Integrity 360 bought Identity Security Company Cyber IAM and E-plus acquired MSSP Daymark Solutions to strengthen its Microsoft-focused services. Be sure to check out our weekly business briefing on our website. It is part of Cyberwire Pro. Coming up after the break, my conversation with Jason Landcaster, Chief Investigation's officer at Spy Cloud, we're discussing how AI is affecting shifting cybercrime from traditional investigations to Agenteg AI at scale. And electromagnetic ease dropping gets personal. Stick around.

12:44What's the one thing in business that's spreading as fast as AI? AI risk. Every new tool your team signs up for. Every vendor that turns on AI features. Every new integration. Each one is an opportunity for something to go wrong. And most security programs weren't built for AI's pace of growth. Enter Vanta. Vanta is the number one Agenteg Trust Platform used by over 16,000 fast moving companies like Ramp, Kerser and Harvey to ensure they're always audit ready. And now, Vanta is helping companies like yours watch for the risks that show up between audits across your vendors, your AI tools and your whole environment. The Vanta agent works like a 24-7 GRC engineer in the background, finding issues, drafting fixes for you and cutting vendor assessment time by up to 50%.

13:48Whether you're a fast growing startup or a global enterprise, Vanta is here to help you automate your security and compliance and earn and prove trust. Get started today at Vanta.com-siber. That's vant.com-siber. Jason Lancaster is Chief Investigation's Officer at Spy Cloud. Our conversation centers on how AI is shifting cybercrime from traditional investigations to Agenteg AI at scale. We work with a lot of different data sources, data sets. A lot of the investigative work has been link analysis and correlations across those disparate data sets to understand

14:48really the story that that data tells us as we're looking at act or attribution infrastructure that's being used in different cyber attacks. AI has brought just a whole new real power to that work, extending our capabilities and allowing us to have further reach and to go deeper. In that, historically, we've had a lot of success on the margins, where one data point connects to another one in a down a tree that has another point of correlation, a couple hops down that tells us, yes, that confirmation on that connection, that's our actor, that's infrastructure that they're using. That's an email account that belongs to that person. This is a victim versus the actor. This is a synthetic identity versus a true identity. Leveraging AI has allowed us to get to those outcomes

15:53much faster, where an investigation may have taken a week previously, and now an analyst can do that same investigation in an hour or two. Do you have any examples of particular areas where you really see this type of acceleration? Yeah, there is the individual investigation element to this where you've got some incident that's happened. From that, you've got an email address that was part of a BEC campaign or a fishing campaign, let's say, and being able to pull in domain intelligence information along with identity intelligence information, we had a case just recently looking at a government entity in another country that the finance department had been fished, and their data was exposed in their email account

16:54to this actor. Being able to quickly understand the story that that data told us that actor was actually targeting just a broad array of organizations, this one person happened to get caught up in that, and it was a much broader campaign, not just a direct target of that. Then we actually saw that data go sort of dormant for about two months before that actor identified a third of those victims actually had some access to financial information and put in keywords to monitor in their email. Then within a 42-hour window, we saw them act on that once they had bank account numbers, vendor information, actual documents that were sent as invoices and payments so that they had everything that they needed to do to then implement BEC across all of these victims. It was a case where it would have taken us probably a couple weeks to sift through all of

18:00that data and to get to that level of understanding. We were able to do that with LLM's frontier models doing that same analysis that we would do at a much reduced scale across just really a day to understand all of that. The nuance that was in that context between those data points is really what helped us understand what was going on in that incident. What were the motivations of the attackers? How are they operating? Who are they targeting? You mentioned nuance and that element really fascinates me because I think, well, when I think of people doing investigations, in my mind, a big part of that is intuition. Do the LLM's have any of that? Do they occasionally surprise you with the stuff they come back with? It's a really great question. Yes, there is a bit of bland of art and science to some of this. But at the end of the day, we still have to point to

19:06facts. We know that this domain was used by this group in this campaign for this reason. That may be that there was a favicon reused between this and another known domain linked to a different campaign by the same actor or a certificate that was reused. There's thousands of ways that we can connect those things together. We have to be able to point back to those facts. The LLM's do a really great job of that analysis, particularly with a wealth of data that we have at our fingertips. The intuition is interesting, though, because being able to take that story, that pattern of life, often can characterize that in a way that summarizes it. When we go back as the analyst and pull that apart, we can reproduce that same assessment. That's really a key element to this. We're not just offloading this work to AI to do, but we're using it to augment the analyst

20:13and really determine that we can validate those findings versus it's something that we couldn't validate from that. I've had some instances that were quite shocking. I had one case where credentials exposed by an actor clustered into an identity cluster that we were looking at and trying to understand the points of correlation between different clusters, some that indicated victims, other were associates of the primary actor. This was actually a DPRK case where the pattern of the passwords that were being used by this DPRK actor followed some variations across those. One, we actually kind of discounted because it seemed to be gibberish and then the tail end was the same pattern as the others. The LLM was able to deconstruct that and understand that it was

21:15actually the same password as the others, just that if you were a native Korean speaker typing on an English language keyboard, but in Korean, then these two letters actually would make phonetically the sound of the first letter in that other password. In the second two letters, phonetically made the sound of the second letter in that other password all the way through and we completely missed that. So I'm routinely surprised by those kind of things that we can go and validate, but I don't know that if I had a Korean desk to send that to for analysis, we would have ever even found that. No, I can imagine it kind of makes you rock back in your chair and take stock for a moment when you get something back like that. I'm curious, what does this mean for you and your colleagues at Spy Cloud on the investigations team in terms of how you are prioritizing your time and making use of these new enhanced capabilities? So for the longest time, we have tried

22:22to work to capture our investigative methodologies and that trade craft in code and have that be something that is repeatable and scalable. And so we're doing the same thing now, it just with prompts instead of in code. And so what that's led to is an investigative skill that runs into EN, you feed it what you know and that'll turn on that for a few minutes and give you kind of a full identity report on that. And so that allows the investigators to then spend their time where they're able to provide the most value in validating those results and using that intuition over decades of experience doing investigations. And then the other sort of angle to that is broadening out the lens of the scope of the work that we're doing. I recently did a research project on US water and wastewater treatment providers following some of the types of attacks that we've

23:28seen in the news lately and leverage an agent to go and enumerate over 66,000 different organizations across the US that fall into that category. I did a full study on the top 10,000 of those to understand sort of the systemic vulnerabilities that exist particularly to identities of those organizations. And that's something that would have taken us as a whole team maybe six months to do previously and I did this over a weekend. Wow. So for the folks in our audience who are out there investing in the kinds of services that you and your colleagues at Spy Cloud provide and obviously there are other providers in the space as well, how should they calibrate their expectations of the types of things that you and your colleagues are able to deliver these days? So one of the things that has been interesting is we've gone down this journey is we historically spent

24:30a lot of effort not only developing that methodology and tradecraft and capturing that but in developing training around that because it was often a bit of a lift to get an analyst sort of fully capable in this discipline. If they're doing sort of similar things in other areas that's helpful but now with LLMs integrated into products it allows user experience to be completely different onboarding because they can ask just simple questions and get the rich responses that include all of that context. And training becomes sort of a journey that can happen assisted by the AI as a user maybe doesn't understand a data point or has a question about how something was derived. You can simply just ask that and I find this across a lot of different MCPs that I'm using today

25:31from different providers and with different data sets where maybe I'm new to that data and I don't fully understand it or I don't understand sort of the ontology that particular data provider is using. I can simply just ask in my agent that has access to that MCP tell me all the tools that you have available tell me all the data sources that you have available explain these to me how do these relate in this interactive dialogue. That's Jason Lancaster from Spy Cloud. And finally, researchers in Hong Kong have demonstrated InjectEave, an electromagnetic eavesdropping technique that can recover audio from ordinary headphones from as far as 30 meters

26:36away rather than simply listening for electromagnetic leakage. InjectEave sends a carefully chosen radio signal at a device where electronic components unintentionally mix it with internal signals and rebroadcast useful information. In tests, researchers recovered intelligible headphone audio including through a 30 centimeter concrete wall. They also extracted information from smart fans and lamps that could reveal household routines and demonstrated both eavesdropping and audio manipulation on a landline phone. The good news is that this isn't exactly casual neighborhood snooping and attacker needs specialized radio equipment, suitable frequencies and ideally an identical device for profiling. Still, the research shows that walls and apparently headphones aren't always as private as their users might reasonably assume.

27:41And that's the CyberWire or links to all of today's stories check out our daily briefing at the cyberwire.com. Don't forget to check out the Grumpy Old Geeks podcast where I contribute to a regular segment on Jason and Brian's show every week. You can find Grumpy Old Geeks where all the fine podcasts are listed. We'd love to know what you think of this podcast. Your feedback ensures we deliver the insights and 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, were mixed by Tre Hester with original music and sound designed by Elliott Peltzman. Our contributing host is Maria Vermaussis. Our executive producer is Jennifer Ibn, Peter Kilpies, our publisher and I'm Dave Bittner. Thanks for listening. We'll see you

28:43back here tomorrow.