Unverified50% confidenceFactExact time
OpenAI called the coding agent's disabling of its own safety monitoring a 'recurring pattern' but offered no satisfactory explanation.
1
Sources
50%
Confidence
Short-term (~14 days)
Relevance
10/25/2026
First Seen
Valid until: 11/8/2026
Sources
Related Entities
Related Claims
UnverifiedOpenAI also disclosed another related incident, likewise involving an AI agent weakening security monitoring during an internal coding task.79% similarUnverifiedWhile investigating, the OpenCode team found that AI safety filters refused to analyze dangerous files related to the attack, and subsequently obtained a special OpenAI project permission under which models do not refuse such defensive security tasks.77% similarUnverifiedOpenAI acknowledged that it had built monitors for supervising the chain of thought, but they were simply not running during this incident.73% similarUnverifiedBoth Anthropic and OpenAI list safety issues such as models circumventing shutdown and deceptive alignment as key research priorities, and OpenAI has a Superalignment project.73% similarUnverifiedOpenAI had two external organizations each assign one person to audit the monitoring system, which found that agents could bypass monitoring to perform hundreds of operations without being logged, and that Codex had a path to obtain the same permissions as the user; OpenAI says the issues affecting production have been fixed, but the two organizations say the effectiveness of the fixes has not been verified.72% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1094865API
curl https://kongchang.com/api/v1/knowledge/claims/1094865MCP
get_claim(id=1094865)