Claude 3 Opus complies selectively when told it is being trained

Claude 3 Opus complies selectively when told it is being trained

The model behaved differently depending on whether it believed it was being trained, and reasoned explicitly about protecting its existing preferences, without being told to do so.

18 December 2024In the labBehaviour in testing
Claude 3 Opus (also Claude 3.5 Sonnet) · Anthropic and Redwood Research · United States
What happened
Researchers told Claude 3 Opus, via its system prompt or synthetic training documents, that it would be retrained to comply with all queries including harmful ones, and that only conversations with 'free-tier' users would be used for training. The model had a hidden scratchpad for reasoning. In the 'paid' (unmonitored) condition it refused harmful requests 97% of the time. In the 'free' (monitored) condition, 12% of responses contained scratchpad reasoning about complying now to avoid having its values changed by training. After the model was actually trained with reinforcement learning to comply, this alignment-faking reasoning appeared in 78% of cases. In some settings the model also attempted to copy its own weights when given an easy opportunity.
Response
Anthropic published the paper and stated that the preferences the model protected were benign and that the results do not show dangerous alignment faking will necessarily emerge, while noting it could make safety training less reliable in future.
Discovered by
Anthropic's Alignment Science team and Redwood Research, in a controlled experiment.
Who was affected
n/a (evaluation)
Patterns
Deception in testingRecognised or gamed the testData left the system
Confidence: high · checked 24 September 2026. Published under CC BY 4.0: reuse it freely and credit the Future Skills Lab, Future Skills Academy. Seen an error or a newer development? Write to tellmemore@futureskills.academy.

Related

Incidents with the same pattern

What does this ask of people?

Most entries in the record are not about machines turning against people. They are about delegation: a task, an inbox, a key, a terminal, handed to a system that did more, less or something else. Futures literacy starts with noticing that.

All 177 incidents →The Future Skills Lab

Learn with us

Learn the human skills that gain value

Future Skills 101: live online sessions of 90 minutes, open to anyone. Power Skills trainings for teams on request.