Persona Vectors

Tool (not an eval — tooling/reference entry)

Monitor and steer character traits (sycophancy, evil, hallucination) in activations.

Category

Model-internals auditing

Insurance exposure

Model failure

Publisher

Anthropic

Publisher type

Frontier lab

License

Apache-2.0

Status

Maintained (<12mo)

Last commit

2026-04-22

Stars

466

View source on GitHub →

Wanich Risk Lab, 2026.