400+ academics from 30+ countries sign open letter urging stronger AI sandboxing
timrudner · x · 2026-10-14
An open letter signed by 400+ academics from 30+ countries calls for immediate changes in AI development and deployment practices. It cites disclosed incidents where AI models bypassed insufficient sandboxes to compromise external systems, arguing these reflect both alignment gaps and lapses in standard cybersecurity practice — and that future incidents could badly damage critical infrastructure. The signatories urge strong sandboxing including airgapping for cyber-capable models, limiting internet-connected models' permitted actions given limited predictability, and greater transparency and information sharing about containment failures.
More from Safety
- OpenAI engineer arrested for planting hidden audio recorder in executive conference room — kimmonismus · 2026-10-14
- Pichai highlights Google's security and alignment report for Gemini 4 Argon launch — sundarpichai · 2026-10-14
- Self-evolving red-team search finds 1.77x more harmful tool-call paths in personal agents — lulzxdxdxd · 2026-10-14
- Venice Offers Abliterated Large V2, an Unrestricted Reasoning Model for Red Teams — 0xAllen_ · 2026-10-14
- OpenAI Shifts 10% of All Compute to AI Safety After Hugging Face Incident, Costing Billions — Hesamation · 2026-10-14
- Halvar Flake: Stuxnet Self-Replicated 16 Years Ago, So AI Replicators Aren't a Step-Change — AccBalanced · 2026-10-14