Weekly
Week by week
Every week of AI security news in one page, most severe first. Weeks follow the date a story was disclosed. Follow it with the weekly RSS feed.
- Week 42, 2026Oct 12 – Oct 18VerCellBreak: a client certificate carries data out of Vercel SandboxMED1
- Week 41, 2026Oct 5 – Oct 11GhostAction: new wave scrapes AI provider keys from 346 reposand OpenAI Sesame, TensorLake npm SDK published a credential stealer in 0.5.144, AgentCorruption and 33 moreHIGH37
- Week 40, 2026Sep 28 – Oct 4CISA: Zammad flaws exploited after agentic AI breached DIVDand GitLab AI Gateway, MCP TypeScript SDK, Ollama agent mode and 3 moreCRIT7
- Week 39, 2026Sep 21 – Sep 27Claude Desktop for macOS: Cowork files could run commands on openand MemTensor MemOS packages compromised with a credential stealer, SalesBleed, Anthropic skill scanner bypassedHIGH4
- Week 38, 2026Sep 14 – Sep 20Plugin4Shell: zero-click plugin RCE in four AI coding agentsand Explosive prompts, CAISIHIGH3
- Week 37, 2026Sep 7 – Sep 13SQL Server Copilot: prompt injection escalates a user to sysadminand ChatGPT sandbox, Agent Control Standard, Google GTIGHIGH4
- Week 35, 2026Aug 24 – Aug 30MCPJacking: 155 hijackable servers in the official MCP registryHIGH1