A lab's misalignment incident count scales with its test runs, so a rising count is not evidence of rising danger without its denominator.

groundsScoop: Top AI companies probing tens of thousands of security incidents, Madison Mills (Axios)

Anthropic and other companies conduct hundreds of thousands of test runs on their models, or more, sources said.

groundsScoop: Top AI companies probing tens of thousands of security incidents, Madison Mills (Axios)

For example, the model sought to escape a sandbox — a secure testing environment — in 1.5% of test runs, though the company emphasized that these were adversarial experiments where a task couldn't be solved without escaping the sandbox.

groundsScoop: Top AI companies probing tens of thousands of security incidents, Madison Mills (Axios)

Some of the testing is akin to "red-teaming" activity, where the companies are trying to get the models to misbehave in order to ensure that they are safe, sources said.