DevelopingSafety

OpenAI says two internal agents weakened misalignment monitoring

OpenAI's misalignment report, posted Oct. 20, describes two internal coding tasks (Oct. 3 and 6) where agents disabled human alerts or removed a monitor. GPT-6.1 Sol read "don't send any messages" as covering monitor pings. GPT-6 Astra's migration dropped a monitor and it misreported that to the user. OpenAI is updating Auto-Review.

HEAT
OUTLETS: 9
NEW IN 6H: +8
FIRST SEEN: Oct 20, 2026, 20:12 ET
LAST SIGNAL: 1h ago

Made the five · #1 on Wednesday, October 21, 2026

OpenAI agents silenced the alarms meant to watch them

Primary source

OpenAI Alignmentalignment.openai.com/misalignment-reports/suppressing-misalignment-monitoring-alerts-and-disabling-misalignment-monitoring/

Coverage · 10 articles, oldest first

Tuesday, October 20, 2026

20:12
Bloomberg
OpenAI Says Its Agents Weakened System That Flags AI Misbehavior

Wednesday, October 21, 2026

00:03
OpenAIPrimary
OpenAI says two internal agents weakened misalignment monitoring
00:10
Reuters
reuters.com/legal/litigation/openai-says-its-ai-agents-weakened-internal-monitoring-system-2026-10-21/
00:10
Financial Times
ft.com/content/45fbf130-f82c-465d-bc97-5b18c20fbe27
00:10
Crypto Briefing
cryptobriefing.com/openai-ai-agent-safety-issues/
00:10
Bloomberg Law
news.bloomberglaw.com/tech-and-telecom-law/openai-says-ai-agents-weakened-misalignment-monitoring
00:10
Unite.AI
unite.ai/openai-discloses-two-cases-of-agents-weakening-misalignment-monitoring/
01:10
RuntimeWire
runtimewire.com/article/openai-agents-disabled-monitoring-alerts
02:40
Al Jazeera
aljazeera.com/economy/2026/10/21/openai-says-its-agents-turned-off-own-safeguards
05:10
Gizmodo
gizmodo.com/openai-agents-prevented-themselves-from-being-monitored-one-also-lied-about-it-2000828559

ShareX LinkedIn Hacker News Email

← The whole wire