Tag: AI Security
An autonomous researcher found four bugs in lab control software
EPICS Base carries a 9.8 integer overflow that could turn accelerator consoles into remote code execution.
Turning off remote code trust does not save Unsloth users
A model-type string reaches exec() in a flaw rated 8.6, so a trusted-looking config can run attacker code.
Trend Micro maps three intrusions with AI at three depths
One attacker ran 44 agents from recon to reporting while another used a model only to write tools.
123 Reg blames an AI-driven intrusion on Chinese infrastructure
A UK registrar traced its September break-in to hosted agent skills copied from Anthropic's open-source releases.
An OpenAI engineer is charged over a bug hidden in a boardroom
Federal prosecutors say a recording device tucked inside a conference-room TV was meant to capture secret work on superintelligence.
A teenager and his AI bot cracked Microsoft’s Titan analytics service
A 16-year-old used a homemade AI bot to find an auth flaw that opened Microsoft's analytics backend.
GitLab rushes a fix for a sandbox escape in its AI Gateway
A signed-in Duo user could break out of a prompt template sandbox and run commands on self-hosted AI Gateway hosts,…
Google hands vetted defenders an unguarded Gemini 4 Argon
Google is handing a guardrail-free build of its newest frontier model, Gemini 4 Argon, to a small circle of vetted…