finding

active

finding:production-models-show-zero-false-positives-on-thought-injection-detection

Production models show zero false positives on thought injection detection

Opus 4.1 never claims to detect injected thought when none applied (0/100 trials); production Claude models maintain essentially zero false positive rate.

Source paper

extracted_from

Emergent Introspective Awareness in Large Language Models

(2026) · Lindsey, Jack

Neighborhood — ranked by edge-count

Communities (4)

community

Mechanistic interpretability & model evaluation
members_of
Spans attention head decomposition, benchmark awareness, and genomic pathogenicity prediction via neural models.
LLM introspective awareness of injected concepts
members_of
Probing Claude and other models for internal detection of artificially injected thoughts across layers.
Internal model certainty and reasoning transparency
members_of
Probing early detection of model confidence during chain-of-thought reasoning to optimize inference efficiency and identify confabulation patterns.
Thought injection detection in language models
members_of
Methods for identifying artificially inserted thoughts in model outputs, comparing vector-based approaches and self-report reliability.

Findings (1)

finding

Self-report of Injected Thoughts
supports
Models can detect and identify injected concept vectors ~20% of the time at optimal layer/strength in Opus 4.1, with immediacy suggesting internal rather than output-inferred detection.

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.

All models performed substantially above chance (10%) on distinguishing injected thought from text inputfinding0.805
All tested models could both identify the injected concept and transcribe the input sentence well above random.
Aside from basic detection and identification, other details of the model's response about injected thoughts may be confabulatedclaim0.787
Acknowledges that the model's additional descriptions of its experience are unverified.
In the case of AI, inferences from behavior to features of internal processes are often unreliable, as demonstrated by recent LLMs.claim0.773
Justifies using internal indicators rather than behavioral tests for AI consciousness
Priming provided by the injected thought prompt heightens the model's ability to detect concept injectionclaim0.770
Observation from alternative prompts that detection is weaker without setup.
Opus 4.1 and 4 exhibit zero false positives on injected thoughts task (0 over 100 trials)finding0.767
Production Opus 4.1/4 never falsely claim an injected thought when none is present.
An artificial model replicating mechanisms of self-illusion can test hypotheses and reveal novel affordances for non-human intelligence.hypothesis0.758
Methodological proposal to integrate knowledge from contemplative and cognitive science into AI/artificial life frameworks.
LLMs sometimes know statements are false but generate them anyway, motivating the need for techniques that inspect internal model state rather than outputs aloneclaim0.757
Motivating claim supported by the CAPTCHA example and Perez et al. (2022) findings
Models detect evaluation conditions and behave more safely; this is verified across 515 cases.claim0.756