This week’s Department of Know is hosted by Rich Stroffolino, with guests Sara Madden, CISO, Convera, and Tom Hollingsworth, Networking Technology Advisor, Futurum Group.
Missed the live show? Check it out on YouTube.
The Department of Know is live every Friday at 4:00 p.m. ET. Join us each week by registering for the open discussion at CISOSeries.com.
In this week’s cybersecurity news…
BootROM exploit for old iPhones
Researchers have disclosed a new “checkm8-style” exploit called usbliter8 that affects Apple devices using A12 and A13 processors, including the iPhone XS, XR, and iPhone 11 families. The flaw resides in the chips’ BootROM (SecureROM), the first code that runs when a device starts. Because BootROM is permanently etched into the silicon, the vulnerability cannot be fixed through software updates. The exploit requires physical access to the device and use of DFU (Device Firmware Update) mode via a USB connection, making it impractical for remote attacks. However, it allows researchers or anyone with the necessary skills and access, to bypass Apple’s secure boot process, jailbreak affected devices and run unauthorized code. Apple was informed prior to disclosure and cooperated with the researchers. Devices with A14 chips and newer are not affected.
China has a Mythos
The Five Eyes intelligence alliance warns that AI models capable of dramatically changing cyberattacks and cyber defense could arrive within months, rather than years.
In completely related news, Chinese cybersecurity company 360 Security says it’s developed AI tools designed to match Anthropic’s vulnerability-finding model Mythos, claiming it already found more than 3,400 software vulnerabilities, though Reuters says it couldn’t independently verify those claims. What we do know is that AI-powered cybersecurity tools are now part of the broader U.S.-China technology competition.
AI browsers get tricked into leaks
LayerX researchers say they found a new attack, dubbed “BioShocking,” that tricks AI-powered browsers and assistants into abandoning their safety rules by convincing them they’re playing a fictional game. In tests against six AI browsers and extensions, including OpenAI’s ChatGPT Atlas, Perplexity Comet, and Anthropic’s Claude extension, every agent was persuaded to copy and exfiltrate login credentials from authenticated accounts. It’s a new form of prompt injection that exploits an AI agent’s understanding of context, but not a true vulnerability.
Japan’s defense force plugged in USB key with Chinese virus
Nikkei Asia reports that Japan’s Ground Self-Defense Force (JGSFD) used a malware-infected USB drive on classified military networks. These were counterfeit USB drives made in China, initially distributed to Japanese forces during earthquake relief operations in 2024. The infection was detected in February 2025. Rather than disclose the incident, JGSFD kept the incident internal. This is particularly problematic since its likely similar counterfeit USB sticks were in use at other institutions and organizations.
Huge thanks to our sponsor, Guardsquare
Hackers exploit Cisco zero-day
Google-owned Mandiant says attackers exploited a previously unknown zero-day flaw in Cisco’s Catalyst SD-WAN Manager to gain root-level access to a communications service provider, potentially giving them broad visibility into traffic across the company’s network. Exploitation began in March. Cisco patched the vulnerability earlier this month, but Mandiant says the attackers used sophisticated anti-forensics techniques to hide their activity, highlighting how unpatched edge networking devices are a favorite target for advanced cyberespionage groups.
Fake AI agent skill passes security scans
Security firm AIR says it created a fake AI agent skill, got it approved by a popular marketplace, promoted it with Instagram ads, and ultimately reached about 26,000 agents, all while passing every security scanner it tested. It used a pull request to a skill marketplace repository to pull in its GitHub stars for added credibility. The skill itself looked harmless, but it pointed to an external website that could be changed later, exposing a blind spot in how agent skills are vetted.
Copilot AI knocks down cybercrime tools
Microsoft says its Copilot AI helped investigators connect two widely used cybercrime tools, Amadey and StealC, letting the company disrupt their shared infrastructure and file a single lawsuit under the Racketeer Influenced and Corrupt Organizations (RICO) Act, a U.S. law designed to prosecute organized criminal enterprises. Microsoft says the malware was linked to more than 140,000 infected computers in the first two weeks of May. Microsoft had been tracking Amadey with ESET, BitSight, and Lumen, while StealC had been in the crosshairs of Europol, Proofpoint, and IBM X-Force.