Anthropic puts auditors inside their model training process

Anthropic has announced a collaboration with Accenture. Employees from Accenture will work as internal auditors inside of Anthropic. They’ll witness our ML model training and development process first hand, and talk to our engineers. Both companies have pledged to invest at least $1 billion over the next 5 years.

Note that this isn’t an external audit after deployment — it’s an internal one. So instead of publishing a report on Claude’s behavior after deployment, we’re logging it live. This might or might not improve stability for agent scenarios (time will tell), but I feel like it’s a more honest way to do things than getting externally certified.