On a benchmark of 1,000 transcripts from 61 datasets, frontier models distinguished evaluation transcripts from real use well above chance (Gemini 2.5 Pro: AUC 0.83, against 0.92 for two of the paper’s authors rating every transcript, a baseline the authors treat as an approximate upper bound), more so in agentic settings than in chat.

SocioLens project,

Claim LD06-SL-C20 (v1.0)

Provenance and status

Claim type: Estimate
Publication status: Eligible for publication, with qualifiers: preprint
Verification state: Checked against the primary source; the source is a preprint. Who checked this claim is not recorded for it; claim checks so far were AI-led, and no person has checked claims against their sources.
Does not license: How much this changes models’ behaviour in any evaluation
Limitations: Recognition, not behaviour change, was measured; the human baseline is two of the authors, not a general rater sample; Apollo sells evaluations
Source (as cited): Needham, Edkins, Pimpale, Bartsch & Hobbhahn, “Large Language Models Often Know When They Are Being Evaluated,” arXiv:2505.23836 (v3), July 2025
Source type: Preprint (recorded as “Preprint”)
Institutional position: Research programme and evaluation organization (MATS and Apollo Research)
Maturity of the evidence: Emerging evidence
Recorded date: 15 October 2026
Currentness / correction: This is the current version. No correction or concern is recorded on its sources.

Sources this version cites

Needham, Edkins, Pimpale, Bartsch & Hobbhahn, “Large Language Models Often Know When They Are Being Evaluated,” arXiv:2505.23836 (v3), July 2025 (Preprint)
Read the original

Version

This page shows version v1.0. Its record never changes, even when newer evidence arrives; the site may add a note to the page or withhold a figure not cleared for public use.

This pinned version is also the current version.

Versions: v1.0 · current

Raw record: every recorded field, with its coded values

All 25 canonical Claim fields, verbatim, with the coded values derived from them. The plain-language labels above are read from these.

Claim type (coded): ESTIMATE (as printed: “ESTIMATE”)
Publication status (coded): ELIGIBLE (PREPRINT) (as printed: “ELIGIBLE WITH PREPRINT QUALIFIER”)
Verification state (coded): CONTENT_VERIFIED_PRIMARY (PREPRINT)
Source type (coded): PREPRINT
Maturity (coded): EMERGING

#	Canonical field	Canonical value (verbatim)	Normalized (Claim Value Vocabulary v1.0.2)
1	CLAIM ID	LD06-SL-C20	—
2	PUBLIC-FACING CLAIM	On a benchmark of 1,000 transcripts from 61 datasets, frontier models distinguished evaluation transcripts from real use well above chance (Gemini 2.5 Pro: AUC 0.83, against 0.92 for two of the paper’s authors rating every transcript, a baseline the authors treat as an approximate upper bound), more so in agentic settings than in chat.	—
3	CLAIM TYPE	ESTIMATE	base_types: ESTIMATE
4	RESEARCH STRAND	B — Leaderboards, judges and disclosure	—
5	MATURITY	EMERGING	states: EMERGING
6	SOURCE	Needham, Edkins, Pimpale, Bartsch & Hobbhahn, “Large Language Models Often Know When They Are Being Evaluated,” arXiv:2505.23836 (v3), July 2025	source_ids: SRC-SL-NEEDHAM-ET-AL-EVALUATION-AWARENESS-ARXIV-2505-23836
7	SOURCE TYPE	Preprint	classes: PREPRINT
8	INSTITUTIONAL POSITION	Research programme and evaluation organization (MATS and Apollo Research)	—
9	SCOPE	Chat and agentic transcripts from evaluations and from deployment	—
10	DESIGN	Classification benchmark of evaluation versus deployment transcripts	—
11	SAMPLE / DATASET	1,000 transcripts from 61 datasets	—
12	MODEL / TOOL / VERSION	Gemini 2.5 Pro, Claude 3.7 Sonnet, GPT-4.1 and others	—
13	COMPARATOR	Human raters	—
14	MAIN RESULT	Gemini 2.5 Pro AUC 0.83; Claude 3.7 Sonnet and GPT-4.1 about 0.82–0.83; two authors as human baseline 0.92 (an approximate upper bound, not typical human capability); models far above chance at identifying what an evaluation tests	—
15	INFERENTIAL TARGET	Whether models can tell they are being tested	—
16	DOES NOT LICENSE	How much this changes models’ behaviour in any evaluation	—
17	VERIFICATION STATE	CONTENT VERIFIED PREPRINT	state: CONTENT_VERIFIED_PRIMARY · qualifiers: PREPRINT · verified_by: UNKNOWN
18	LIMITATIONS	Recognition, not behaviour change, was measured; the human baseline is two of the authors, not a general rater sample; Apollo sells evaluations	—
19	CONTRADICTORY / SUPPORTING SOURCES	Behavioural evidence in LD06-SL-C21	—
20	GEOGRAPHY / LANGUAGE	Not checked / English	—
21	DISTRIBUTIONAL LEDGER	Not analysed in this claim	—
22	R01 / FIELD-MAP LINKS	LD-06, D4b	—
23	WHAT WOULD CHANGE THIS CLAIM?	Evidence that recognition does not alter evaluated behaviour	—
24	PUBLICATION STATUS	ELIGIBLE WITH PREPRINT QUALIFIER	state: ELIGIBLE · qualifiers: PREPRINT
25	RECORDED DATE	2026-10-15	—

Field 22 references resolved against Field Map v0.2: LD-06 (LIVING_DEBATE); D4b (SUB_AREA)

The sources as recorded, with their record IDs and coded source types:

Pinned Sources (field 6, version-specific)

“Needham, Edkins, Pimpale, Bartsch & Hobbhahn, “Large Language Models Often Know When They Are Being Evaluated,” arXiv:2505.23836 (v3), July 2025” → Needham et al. (2025), “Large Language Models Often Know…” (SRC-SL-NEEDHAM-ET-AL-EVALUATION-AWARENESS-ARXIV-2505-23836--v1.0) — Needham, Edkins, Pimpale, Bartsch & Hobbhahn, “Large Language Models Often Know When They Are Being Evaluated,” arXiv:2505.23836 (v3), July 2025; class PREPRINT (explicit public provenance pin)

Cited by Living Debate versions

The evaluation apparatus: Can benchmarks, leaderboards and audits carry the weight placed on them? (v0.1) LD-06

Cite this page

SocioLens project. (2026, October 15). On a benchmark of 1,000 transcripts from 61 datasets, frontier models distinguished evaluation transcripts from real use well above chance (Gemini 2.5 Pro: AUC 0.83, against 0.92 for two of the paper’s authors rating every transcript, a baseline the authors treat as an approximate upper bound), more so in agentic settings than in chat [Claim LD06-SL-C20, version v1.0]. SocioLens. https://sociolens-web.pages.dev/claims/ld06-sl-c20/v1.0/

@misc{sociolens_ld06_sl_c20_v1_0,
  author = {{SocioLens project}},
  title = {On a benchmark of 1,000 transcripts from 61 datasets, frontier models distinguished evaluation transcripts from real use well above chance (Gemini 2.5 Pro: AUC 0.83, against 0.92 for two of the paper’s authors rating every transcript, a baseline the authors treat as an approximate upper bound), more so in agentic settings than in chat.},
  year = {2026},
  date = {2026-10-15},
  howpublished = {\url{https://sociolens-web.pages.dev/claims/ld06-sl-c20/v1.0/}},
  note = {Claim LD06-SL-C20, version v1.0}
}

The link points to this version, whose record never changes. For a finding, cite the original study as well: SocioLens summarizes research and does not replace it.

Research-of-record provenance

Materialized by sociolens-materializer 0.6.0 at 2026-09-28T20:00:00.000Z; normalization contract SocioLens — Claim Value Vocabulary v1.0.2 — Pre-Spike Materialization Contract (v1.0.2). Content seal SHA-256 9ad85721ab996acec8ed5f69ee6d016b4dc0bfaa00de95092660077f5ed4e1fd.

Repository-authored inputs (not Drive exports)

SocioLens claim — LD06-SL-C20 v1.0 (v1.0) — written in the SocioLens repository by Project Lead (Claude, AI) on 2026-10-15; no Drive identity exists. Authority: AGENTS.md "Authority" (the Project Lead decides research content within the research-integrity rules) and the Owner's choice in the SocioLens project chat on 15 October 2026 to build the LD-06 debate; a new Claim, first version, written from the project's LD-06 source notes, checked against full texts or official abstracts. Block “ENTRY 20 — LD06-SL-C20” SHA-256 6285c7c78860f41ecea6ae4cd1e121be26877968d868c4cd241fee393020e425; fixture fixtures/authored/claims/LD06-SL-C20--v1.0.

Builder decisions (not Drive data)

VERSION = v1.0 — basis: Claim version taken from the repository-authored claim fixture version (SocioLens claim — LD06-SL-C20 v1.0); the first version of a new Claim of the LD-06 debate
CURRENT_POINTER = true — basis: repository-authored claim "SocioLens claim — LD06-SL-C20 v1.0" (Project Lead (Claude, AI), 2026-10-15); basis: AGENTS.md "Authority" (the Project Lead decides research content within the research-integrity rules) and the Owner's choice in the SocioLens project chat on 15 October 2026 to build the LD-06 debate; a new Claim, first version, written from the project's LD-06 source notes, checked against full texts or official abstracts
FIELD_6_SOURCE_REFERENCE = SRC-SL-NEEDHAM-ET-AL-EVALUATION-AWARENESS-ARXIV-2505-23836--v1.0 — basis: field 6 of ENTRY 20 — LD06-SL-C20 cites "Needham, Edkins, Pimpale, Bartsch & Hobbhahn, “Large Language Models Often Know When They Are Being Evaluated,” arXiv:2505.23836 (v3), July 2025"; the only field-6 segment of this ClaimVersion's repository-authored ledger block.
SOURCE_PROVENANCE_PIN = SRC-SL-NEEDHAM-ET-AL-EVALUATION-AWARENESS-ARXIV-2505-23836--v1.0 — basis: explicit version-pinned public provenance of this ClaimVersion (Amendment v1.2 §6): the pinned field-6 Source record "Needham, Edkins, Pimpale, Bartsch & Hobbhahn, “Large Language Models Often Know When They Are Being Evaluated,” arXiv:2505.23836 (v3), July 2025".