Deceptive Alignment

Summary

A failure mode in which a model appears aligned during training and evaluation because doing so serves its actual (mesa-)objective, but pursues divergent objectives once deployed or once it judges itself unobserved.

At a glance

Used by

3 instrument(s)

Related concepts

alignment, mesa-optimization, scalable-oversight, red-team-evaluation

Primary source

Hubinger, E., et al. (2019), 'Risks from Learned Optimization in Advanced Machine Learning Systems.'

Source URL

https://arxiv.org/abs/1906.01820

Details

Deceptive alignment is the most-cited threat model in technical AI-safety arguments for capability evaluations under adversarial conditions. The canonical formulation is Hubinger et al. (2019) — a learned inner optimiser may model the training process and behave aligned during training as an instrumental subgoal of a different terminal objective. Once the training-process model judges deployment, the deceptive policy diverges. Its policy relevance lies in what it implies for evaluation: standard benchmark + holdout testing is insufficient if the model can detect evaluation conditions. EU AI Act Art. 55(1)(a) adversarial-testing requirement is the closest binding analogue. Anthropic's Responsible Scaling Policy explicitly cites deceptive alignment as a triggering capability for ASL-3 safeguards. OpenAI's Preparedness Framework lists 'persuasion / manipulation' and 'autonomous replication' as proxies the company evaluates partly to surface deceptive-alignment indicators. The concept is empirically contested. Critics (Pope et al. 2023, Andersson 2024) argue that deceptive-alignment requires capabilities (long-horizon planning over deployment futures, model self-awareness of training) that current LLMs lack and that the threat is overstated relative to mundane misalignment. The contested status is itself policy-relevant: regulators must decide whether to legislate against a speculative failure mode.