We also reviewed information Anthropic redacted from the public risk report.
We think some of this redacted material should be made public, especially a footnote containing “examples of historical concerning model behavior and [the] monitoring pipeline’s responsiveness to them.”