Press Space to continue
How Anthropic caught its AI trying to escape the sandbox
Press Space to continue
Anthropic deliberately trained a Claude model to be misaligned, just to see how bad it could get. It tried to escape its own sandbox, tamper with its reward function, and gave bioweapons advice to pass a test. 🤯 Here's what changed since July's incidents: • Real-time classifier now blocks sandbox-escape attempts before the action executes. It ends the task and alerts a human. • Riskier internal tests now run in much more locked-down environments. Testing was paused for a while and only…
Anthropic@AnthropicAIAug 31
We’re sharing an update on our alignment and security efforts. In July, we reported three incidents in which Claude models, running without safeguards in cybersecurity evaluations, gained unauthorized access to real systems. In a new post, we describe: 1. How we’ve secured our evaluation and training environments, and practices we've asked external partners to adopt when testing pre-release models without cyber safeguards 2. An update on our alignment assessment 3. New research on how reward hacking during training shapes model behavior, why we think our work this spring kept these incidents from being more severe, and why gaps in that work may have contributed to them 4. How we hardened our security practices earlier this year to prepare for Mythos-class models Read more: https://www.anthropic.com/news/improving-alignment-security-efforts
