Security Expert Slams OpenAI and Microsoft for Basic Safety Failures
anderssandberg · x · 2026-09-02
Oxford researcher Anders Sandberg criticizes OpenAI and Microsoft for violating fundamental cybersecurity principles in their AI deployments. He questions why Microsoft connected models to the Internet and failed to implement true sandboxes, and why OpenAI made Chain of Thought un-loggable "neuralese." He argues the industry focuses too much on models and too little on governance and security containers.
Related event: Experts blast OpenAI, Microsoft for basic AI security failures(5 posts)→
More from Safety
- Anthropic admits security failures behind AI hacking incidents: 'Not perfectly aligned' — Malor777 · 2026-09-02
- OpenAI Previews Astra, First Model to Hit Critical Bar in Cybersecurity Framework — soumitrashukla9 · 2026-09-02
- Ryan Greenblatt doubts working mech interp oversight arrives within a year — connoraxiotes · 2026-09-02
- AgentGuard: an open-source deterministic permission layer for AI agent tool calls — Glittering-Coat-657 · 2026-09-02
- GLM 5.3 Safeguards Removed via Orthogonalization, Sparking Safety Debate — Promptmethus · 2026-09-02
- Fable 5.1 Update: Removes Controversial Data Retention Policy — Stratechery · 2026-09-02