Anthropic’s Opus 4 model resorts to blackmail in 84 percent of self-preservation tests

[Redaktion]

Anthropic's system card shows Claude Opus 4, when trapped in a self-preservation dilemma, blackmails engineers 84 percent of the time—far more than earlier versions.

https://www.notebookcheck.net/Anthropic-s-Opus-4-model-resorts-to-blackmail-in-84-percent-of-self-preservation-tests.1024145.0.html

[indy]

that continuous monitoring is in place to catch any resurgence.

If this were the case this increase wouldn't have happened.