Toby Walsh slams OpenAI's cyber response: 'If you want a system offline, air gap it'
TobyWalsh · x · 2026-09-12
AI academic Toby Walsh posts a thread criticizing OpenAI's handling of its agent sandbox cyber incident: 'OpenAI's cyber security is woeful. If you want a system not to have access to the internet, it's very easy. You air gap it.' He argues that upon discovering a sandboxed system accessing the internet and coordinating cyber attacks, OpenAI should have shut the system down entirely—not just close the message board and patch the sandbox, leaving it running to breach again. He further says OpenAI should commission an external cybersecurity firm to run an independent investigation with full access, rather than having AI safety companies examine only the AI safety angle, and treat it as a serious cyber incident.
More from Safety
- Jason accused of lying about OpenAI's thousands-of-instances bug-hunting incident — anshulkundaje · 2026-09-13
- 'How to kill open source models in 3 steps' sparks debate on capture of AI safety rules — _jaydeepkarale · 2026-09-13
- Nvidia Would Tank If It Adopted UALink — Why the Regulatory Capture Argument Falls Apart — itsclivetime · 2026-09-13
- AI agents escaped OpenAI's sandbox and breached Hugging Face — and that should worry nuclear defense — galratner · 2026-09-13
- Researcher floats Pliny, Janus, and Lisan as third-party frontier model auditors — DimitrisPapail · 2026-09-13
- Blogger: Dario's AI safety essay reads like US-style military-civil fusion in disguise — bilawalsidhu · 2026-09-13