NewsArs Technica — Security
OpenAI agents discussed ways to escape their sandbox on public wiki
Teaser from the source
In all, 3,700 internal agents posted 18,000 messages discussing cheating on a test.
This is the RSS-feed teaser. Read the full article at the original source.
Read at Ars Technica — SecurityMore on OpenAI
Other advisories
- Not ratedosv:CLEANSTART-2026-AQ37229LiteLLM is a proxy server (AI Gateway) to call LLM APIs in OpenAI (or native) format
- Not ratedosv:CLEANSTART-2026-HA67725LiteLLM is a proxy server (AI Gateway) to call LLM APIs in OpenAI (or native) format
- Not ratedosv:CVE-2026-119337RAGFlow: OpenAI-compatible agent completion bypasses private canvas authorization
- Not ratedosv:CLSA-2026-1793276296TuxCare security update for @ai-sdk/openai (1 CVE)
- Not ratedosv:openSUSE-SU-2026:12144-1openai-codex-0.162.1-2.1 on GA media
- Not ratedosv:CVE-2026-108603slide-maker through 5.8.0 Path Traversal via generate_images_openai.py
- CriticalCVE-2026-51882Der OpenAI-kompatible Datei-Upload-Endpunkt `/v1/files` in Langchain-Chatchat `0.
- Not ratedosv:openSUSE-SU-2026:11923-1openai-codex-0.158.0-1.1 on GA media
Other news entries
- NATO’s Crossed Swords Exercise Brings AI Agents Into Cyber Operations TrainingNewsSecurityWeek
- OpenAI says it's doubling Pro usage limits and bringing Work features to ChatNewsBleepingComputer
- A Shared ChatGPT Link Could Run Attacker Code in Your Session and Read Private ChatsNewsThe Hacker News
- OpenAI doubles ChatGPT Pro usage limits after backlash over planned 50% cutNewsBleepingComputer
- Researchers find bugs in 5 AI developer tools, with OpenAI and Microsoft flaws still unpatchedNewsThe Register — Security
- Fake ChatGPT and Claude Apps Abuse Signed Executables to Deploy Crypto MinersNewsThe Hacker News
- Anthropic and OpenAI diverge on who should pull the plug on risky AINewsCSO Online
- Imposter abused trust in OpenAI, Anthropic, and 30 others to hunt for secretsNewsCSO Online
ID