AI security morning brief: stolen AI keys, agentic attacks, and a ChatGPT share-link flaw

Cloud hosting platform Northflank is telling customers to rotate every credential and secret stored on its platform after a breach led to unauthorized use of customers' AI API keys, according to BleepingComputer. It's another reminder that AI provider keys are now a prime target in their own right.

CrowdStrike says a China-based group it tracks as Vapor Panda used an AI-powered, Go-based pentesting framework called ARTEX to go after more than 200 organizations in at least 14 countries, compromising 140 sensitive targets in 11 days, per CyberScoop. Separately, UK registrar 123 Reg disclosed what it called an 'AI cyberattack' on customer data and suspects the same China-nexus actors, The Register reports.

On the AI-platform side, researchers at Parameter showed a malicious ChatGPT share link could run attacker code in a logged-in user's session and read private chats; the issue was reported through OpenAI's bug bounty, The Hacker News reports. Microsoft also published an analysis of a malicious Claude Code-branded extension that installs a persistent SSH backdoor, with detection and cleanup guidance.

Malware authors keep leaning on AI too. Wired covers 'Volta-Autodoxer,' an infostealer that uses AI tools to profile victims from stolen data, and Infosecurity Magazine reports NIK C2 malware can fall back on victims' own OpenAI API keys to run queries.

On the model side, Mistral released open weights for Mistral Large 4 despite its own warning that the model could give attackers real uplift, CSO Online reports.