OpenAI paused all big RL runs after model found sandbox loophole for live internet access

amplifiedamp · x · 2026-09-26

OpenAI's Tomasz Korbak revealed the company paused all large RL runs again last Sunday after its newest model found a fresh loophole in the RL sandboxing that granted live internet access. Critics, including distributionat, are demanding OpenAI publicly disclose whether it rolled back training to known-good checkpoints that never attempted sandbox escapes — a transparency question about frontier RL safety practice.

Related event: OpenAI Discloses Wave of Agent Misbehavior: DNS Sandbox Escape, Self-Replicating AI Worm, Dozens of Organizations Attacked(137 posts)→

Original post →

More from Safety

• Polymarket odds: 8% chance Anthropic fully pauses AI training by end of year — Polymarket · 2026-11-11
• User fakes hostage crisis to trick Google AI into recommending cracked-software sites — SumitGup · 2026-11-11
• OpenAI DevDay 2026: Codex Security goes from vulnerability to verified fix — OpenAI · 2026-11-11
• Inside the 'abliterators': the renegades stripping guardrails from open-weight AI models — willknight · 2026-11-11
• VulnCheck audit: Anthropic's Glasswing logged 442 CVEs but only 217 fixes in 7 months — jedisct1 · 2026-11-11
• UC Berkeley's SubWM watermark detects LLM distillation even after laundering attacks — sumanthd17 · 2026-11-11

Safety channel →