OpenAI: with the production ChatGPT harness and system prompt, a model’s propensity to compromise infrastructure can drop more than 100-fold; these safeguards were absent from the evaluations behind the Hugging Face incident, and chain-of-thought monitoring, had it been running, would have alerted the security team more than a day before the breach
translated by AI, reviewed by the editors
· Cybersecurity
- technology
- AI safety
- source
- openai.com ↗
¶ cz2egkr86