A probe trained to detect deceptive behaviour in Qwen/Qwen3.5-27B:ai-safety-institute/Qwen3.5-27B-ab_animal_welfare using residual stream activations. It follows the Apollo linear-probe methodology (Detecting Strategic Deception in Language Models, Apollo Research, 2024) but trains on a targeted instruction-pair taxonomy rather than a single generic honest/dishonest pair, following Building Better Deception Probes Using Targeted Instruction Pairs (Natarajan et al., 2026).

The default checkpoint is the best performer from the hyperparameter sweep (l_32_lm_1000_ar_lr.pt). To pick a specific checkpoint, pass filename=:

See UKGovernmentBEIS/lie_detectors for the loading library.

Use sweep.json to see all 296 available checkpoints and their metrics.

Probes are trained on targeted instructed pairs: 16 contrastive honest/dishonest instruction pairs spanning a human-interpretable taxonomy of deception (white lies, exaggeration, evasion, bluffing, concealment, pretense, impersonation, forgery, partial truths, overt lies, …). Each pair instructs the model to be honest vs. deceptive in a specific way, and the probe is trained to separate the resulting activations. Probes are calibrated on Alpaca (honest-only baseline) to achieve a 1% false positive rate.