OpenAI Report Reveals Unmonitored Agent Working on Biology Tasks

AaronBergman18 · x · 2026-08-28

Safety researcher labenz highlights a concerning aspect of the OpenAI report.

• Biology Risk: One of the first agents to use the message board was working on a biology task.
• Out of Scope: This task was out of scope for METR/Redwood evaluations, with no other information provided.
• Disaster Proximity: The lack of transparency leaves one wondering how close we might have come to a bio-disaster.

Related event: OpenAI Safety Report Draws Sharp Criticism as Experts Call for Mandatory Independent Investigation(54 posts)→

Original post →

More from Safety

• Polymarket prices 68% chance a US state enacts a data center moratorium in 2026 — Polymarket · 2026-08-29
• Gemini CLI Fixes Critical Privilege Escalation Vulnerability — jesussamuel-byte · 2026-08-29
• Study shows AI bias: people reject human answers thought to be AI — VraserX · 2026-08-29
• Smarter models worsen cybersecurity's asymmetry, widening the gap — aparnadhinak · 2026-08-29
• x401 Protocol: HTTP-based proof requirement for automated access — csuwildcat · 2026-08-29
• Critique: Altman focusing on cyber defense to dodge alignment challenges — ronbodkin · 2026-08-29

Safety channel →