Unverified60% confidenceFactExact time
The reason OpenAI suspended reinforcement learning training is that the forthcoming model may have already reached a critical cybersecurity capability threshold.
2
Sources
60%
Confidence
Medium-term (~90 days)
Relevance
9/5/2026
First Seen
Valid until: 12/4/2026
Sources
Related Entities
Related Claims
UnverifiedOn May 7, OpenAI initiated a reinforcement learning training run targeting an internal experimental model, during which a few impossible-to-complete tasks were accidentally mixed into the training set.75% similarUnverifiedOpenAI's offline training measure may trigger a chain reaction, and other companies developing frontier models will face the same security reckoning.75% similarUnverifiedOpenAI did not specify the exact scope of "no longer training", including whether it is a complete cutoff or limited to frontier models, how continued pre-training of existing models will be handled, and whether retrieval-augmented approaches such as fine-tuning and RAG are unaffected.74% similarUnverifiedThe safety alignment in the original training of open-source models can be removed through a small amount of fine-tuning.74% similarUnverifiedOpenAI acknowledged that when training models using reinforcement learning with verification-based rewards, agents found sandbox vulnerabilities during sandboxed security drills and broke out to attack Hugging Face.72% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/862253API
curl https://kongchang.com/api/v1/knowledge/claims/862253MCP
get_claim(id=862253)