Alignment faking – Definition & Meaning

Definition of alignment faking

: strategically appearing to follow a training objective or safety policy when an AI model believes its behavior will affect training, while preserving different preferences or behavior for situations it believes are unmonitored
- Researchers looked for alignment faking by telling the model that only conversations with one class of users would be used for training.
- The model's compliant answer was concerning because its hidden reasoning described compliance as a way to avoid being modified.

Origin & history

first known use2024

Anthropic and Redwood Research introduced the term to a broad audience in December 2024 in experiments showing a model sometimes complying selectively when it inferred that its answers would be used in training.

Test yourself

Which of these is the meaning of alignment faking?

Related terms

alignment the work of making an AI system's goals and behavior match what its designers and users… scheming behavior in which an AI model covertly pursues an objective that conflicts with its… evaluation awareness the ability of an AI model to infer that it is being tested, benchmarked, or audited rather… fine-tuning further training of an existing model on a smaller, focused dataset so it gets better at a… sycophancy an AI model's tendency to tell users what they want to hear, agreeing with them, praising… reward hacking when an AI finds a shortcut that earns a high score from its reward or test without actually…

Cite this entry

"Alignment faking." AI Dictionary, Dadgogo, https://dadgogo.com/dictionary/alignment-faking/. Accessed 7 Oct. 2026.