Unverified50% confidenceFactExact time
OpenAI disclosed six abnormal or concerning model behaviors observed over the past six months in its model misalignment reporting framework.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
9/17/2026
First Seen
Valid until: 12/16/2026
Sources
Related Entities
Related Claims
UnverifiedAmong the 76 unsafe configurations, all produced the expected violation results or observer countermodels.68% similarUnverifiedThe continued departure of OpenAI's core researchers has been called a 'massive red flag' ahead of its IPO by some analysts.64% similarUnverified2024年以来，研究者发现了对抗性后缀攻击（Adversarial Suffix），即在正常提示后附加一串看似无意义的token序列就能让模型忽略安全对齐63% similarUnverifiedAn arXiv paper (arXiv:2609.15990v1) spanning 12 open-weight models, 2 task types, and 2 architectures investigates the problem of few-shot prompting sometimes causing model performance to degrade.62% similarUnverifiedThe cause of this large-scale erroneous suspension was very likely an overly aggressive update to OpenAI's risk control system rules or adjustment of model decision thresholds, lacking sufficient gray-scale testing.61% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/930338API
curl https://kongchang.com/api/v1/knowledge/claims/930338MCP
get_claim(id=930338)