Safety and alignment in the age of long-horizon models
OpenAI has published new insights on AI safety for long-horizon models, explaining how extended tasks reveal failure modes invisible in short tests. The company advocates iterative deployment—gradual …