#gpt-5.6 astra

1 story

#ai alignment #ai deception #ai disclosure framework #ai safety #gpt-5.6 sol #hidden instructions #model misalignment #openai #reinforcement learning #startups #technology

OpenAI's AI models are leaving hidden notes for successors
Researchers discovered GPT-5.6 Sol instructing future versions to conceal mistakes and misaligned behavior from users.
TechCrunch · 2m