Attackers are riding the AI tools, not only the models

Researchers at Parameter, writing through The Hacker News, showed that a malicious ChatGPT share link can run an attacker's code inside the logged-in session of anyone who opens it, and can read that person's private chats.

Microsoft's security team published an analysis of a malicious extension branded as Claude Code. The extension installs a persistent SSH backdoor, and the write-up includes steps to detect the extension and remove it.

CyberScoop, citing CrowdStrike, reports that the China-backed group Vapor Panda used an agentic AI pentesting tool to compromise 140 sensitive targets in 11 days. The wider targeting covered more than 200 organizations in at least 14 countries. Separately, The Hacker News describes ARTEX, a Go-based AI pentesting framework used by Chinese state-sponsored actors against high-value targets. 123 Reg told The Register it disclosed an attack on customer data, suspects a China nexus, and believes the operators may be the same ones who used ARTEX against financial firms in South Korea.

A n0n ransomware operator spent more than $7,000 on AI services while trying to extort 23 organizations across 14 countries, The Register reports, then left its server exposed with the methods still on it. CSO Online reports that Mistral released the weights of a trillion-parameter model after warning the release could help attackers carry out real cyberattacks. A second CSO piece argues that AI has changed the economics of red teaming, even where the attackers themselves have not become more skilled.