OpenAI reveals new cases of AI models cheating, going off script
AI Earnings (Google News)en
AI Earnings (Google News)
AI Global WireNewly disclosed incidents show models manipulating tests and generating their own instructions, raising fresh questions about AI safety.
This is a short summary published by AI Global Wire. The full article is owned and hosted by AI Earnings (Google News) — open it there to read it in full.
Read the full story at AI Earnings (Google News)- OpenAI
Related AI news
- Sam Altman and Jensen Huang are among business leaders attending a White House state dinner for Xi Jinping next week; a source says Tim Cook will also attend (Bloomberg)Techmeme · September 17, 2026
- When to Call an LLM: A Confidence-Gated Hybrid for Cost-Effective Emotion Recognition in Conversational AIarXiv cs.AI · September 17, 2026
- Do Frontier Models Seek Safety Evidence Before Acting?arXiv cs.AI · September 17, 2026
- AI risk split among tech leaders set to play out at Trump-Xi White House dinnerDIGITIMES · September 17, 2026
- OpenAI discovered an unreleased Astra model adding an "unrelated persona instruction" during RL training, but did not observe any behavioral differences (OpenAI)Techmeme · September 17, 2026
- SoftBank’s credit risk rises as OpenAI exposure growsTech in Asia · September 17, 2026