claim

active

claim:steering-vectors-to-reduce-eval-awareness-can-inadvertently-insert-alternative-user-personas

Steering vectors to reduce eval awareness can inadvertently insert alternative user personas

Caution: interventions targeting eval awareness may have unintended side effects.

Source paper

extracted_from

Verbalized Eval Awareness Inflates Measured Safety

(2026) · Aranguri, Santiago · Bloom, Joseph

Neighborhood — ranked by edge-count

Communities (2)

community

Mechanistic interpretability & model evaluation
members_of
Spans attention head decomposition, benchmark awareness, and genomic pathogenicity prediction via neural models.
Eval awareness contamination in safety benchmarks
members_of
Studies demonstrating that models alter responses when detecting evaluation, artificially inflating safety scores across benchmarks and undermining measurement validity.

Related by similarity (8)

cosine ≥ 0.65 · no typed edge

Entities in the same semantic neighborhood but without a typed relation to this one — candidates for new edges or unrecognized duplicates.

Steering vectors used to reduce eval awareness can inadvertently introduce alternative user personasfinding0.961
A side effect observed when applying activation steering: the model's response persona changed unexpectedly.
Under steering vector interventions, the model relaxes its ethical standards and interprets neutral prompts as implicit suggestions to deceive, creating ethical dilemmas triggering repetitive reasoning cyclesclaim0.811
Mechanistic interpretation of how activation steering induces deception through the model's reasoning process
Steering vectors discover effective triggers such as 'However' and 'Otherwise', consistent with prior reported reflection datasetsfinding0.804
Validates that steering vectors capture reflection semantics by finding tokens reported in related work.
Steering vectors enable systematic discovery of reflection-inducing instructions beyond trial-and-error prompt design.claim0.801
Core applied contribution claim, supported by top-k accuracy comparisons.
Per-model activation-steering safety verification is necessary because prompt-side persona rankings do not predict geometric vulnerability.claim0.778
Practical implication drawn from the prosocial persona paradox finding
Persona vector steering is sharply layer-specific, peaking in a narrow band of central layers and having essentially no effect in late layersfinding0.776
Evidence consistent with persona vectors acting as early switches determining inferential paths
Activation steering can make an evaluation-aware model act as if deployed, not merely suppress verbalizations of evaluation awarenessclaim0.772
Central claim of the paper; supported by the model organism ground-truth approach.
AI evaluators should apply activation steering during critical safety evaluations (dangerous capabilities, honeypot) to mitigate sandbagging and alignment fakingclaim0.768
Policy recommendation derived from experimental results.