{"id":"https://openalex.org/W7204185906","doi":"https://doi.org/10.48550/arxiv.2608.21766","title":"Evaluation Awareness in Language Models: Representation, Verbalization, and Control","display_name":"Evaluation Awareness in Language Models: Representation, Verbalization, and Control","publication_year":2026,"publication_date":"2026-08-22","ids":{"openalex":"https://openalex.org/W7204185906","doi":"https://doi.org/10.48550/arxiv.2608.21766"},"language":"en","primary_location":{"id":"pmh:oai:arXiv.org:2608.21766","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2608.21766","pdf_url":"https://arxiv.org/pdf/2608.21766","source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"host_organization_lineage_names":[],"type":"repository"},"license":null,"license_id":null,"version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},"volume":null,"type":"preprint","is_traditional":true,"indexed_in":["arxiv","datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://arxiv.org/pdf/2608.21766","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5063965257","display_name":"Farzaneh Heidari","orcid":"https://orcid.org/0000-0002-4020-525X","observed_orcids":["https://orcid.org/0000-0002-4020-525X"]},"institutions":[{"id":"https://openalex.org/I4210164802","display_name":"Mila - Quebec Artificial Intelligence Institute","ror":"https://ror.org/05c22rx21","country_code":"CA","type":"facility","lineage":["https://openalex.org/I4210164802"]},{"id":"https://openalex.org/I70931966","display_name":"Universit\u00e9 de Montr\u00e9al","ror":"https://ror.org/0161xgx34","country_code":"CA","type":"education","lineage":["https://openalex.org/I70931966"]}],"countries":["CA"],"is_corresponding":false,"raw_author_name":"Heidari, Farzaneh","raw_affiliation_strings":["DIRO , Universit\u00e9 de Montr\u00e9al","Mila -Quebec AI Institute"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"DIRO , Universit\u00e9 de Montr\u00e9al","institution_ids":["https://openalex.org/I70931966"]},{"raw_affiliation_string":"Mila -Quebec AI Institute","institution_ids":["https://openalex.org/I4210164802"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5116870548","display_name":"Amin Memarian","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I4210164802","display_name":"Mila - Quebec Artificial Intelligence Institute","ror":"https://ror.org/05c22rx21","country_code":"CA","type":"facility","lineage":["https://openalex.org/I4210164802"]}],"countries":["CA"],"is_corresponding":false,"raw_author_name":"Memarian, Amin","raw_affiliation_strings":["Mila -Quebec AI Institute"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Mila -Quebec AI Institute","institution_ids":["https://openalex.org/I4210164802"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5023766963","display_name":"Guillaume Rabusseau","orcid":"https://orcid.org/0000-0001-8090-2810","observed_orcids":["https://orcid.org/0000-0001-8090-2810"]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Rabusseau, Guillaume","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":2,"corresponding_author_ids":[],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":null,"has_fulltext":true,"cited_by_count":3,"citation_normalized_percentile":null,"cited_by_percentile_year":null,"biblio":{"volume":null,"issue":null,"first_page":null,"last_page":null},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10028","display_name":"Topic Modeling","score":0.37229999899864197,"subfield":{"id":"https://openalex.org/subfields/1702","display_name":"Artificial Intelligence"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},"topics":[{"id":"https://openalex.org/T10028","display_name":"Topic Modeling","score":0.37229999899864197,"subfield":{"id":"https://openalex.org/subfields/1702","display_name":"Artificial Intelligence"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},{"id":"https://openalex.org/T11689","display_name":"Adversarial Robustness in Machine Learning","score":0.26489999890327454,"subfield":{"id":"https://openalex.org/subfields/1702","display_name":"Artificial Intelligence"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},{"id":"https://openalex.org/T12026","display_name":"Explainable Artificial Intelligence (XAI)","score":0.2151000052690506,"subfield":{"id":"https://openalex.org/subfields/1702","display_name":"Artificial Intelligence"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}}],"keywords":[{"id":"https://openalex.org/keywords/evaluation-awareness","display_name":"evaluation awareness","score":0.9980000257492065},{"id":"https://openalex.org/keywords/language-models","display_name":"language models","score":0.8270000219345093},{"id":"https://openalex.org/keywords/verbalization","display_name":"verbalization","score":0.5299999713897705},{"id":"https://openalex.org/keywords/steering","display_name":"steering","score":0.6039999723434448},{"id":"https://openalex.org/keywords/olmo","display_name":"OLMo","score":0.5929999947547913},{"id":"https://openalex.org/keywords/supervised-fine-tuning","display_name":"supervised fine-tuning","score":0.7129999995231628},{"id":"https://openalex.org/keywords/llm-as-a-judge","display_name":"LLM-as-a-judge","score":0.6150000095367432},{"id":"https://openalex.org/keywords/safety-benchmark","display_name":"safety benchmark","score":0.3959999978542328}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.6151999831199646},{"id":"https://openalex.org/C2775924081","wikidata":"https://www.wikidata.org/wiki/Q55608371","display_name":"Control (management)","level":2,"score":0.611299991607666},{"id":"https://openalex.org/C2777267654","wikidata":"https://www.wikidata.org/wiki/Q3519023","display_name":"Test (biology)","level":2,"score":0.5325999855995178},{"id":"https://openalex.org/C137293760","wikidata":"https://www.wikidata.org/wiki/Q3621696","display_name":"Language model","level":2,"score":0.5254999995231628},{"id":"https://openalex.org/C180747234","wikidata":"https://www.wikidata.org/wiki/Q23373","display_name":"Cognitive psychology","level":1,"score":0.5200999975204468},{"id":"https://openalex.org/C15744967","wikidata":"https://www.wikidata.org/wiki/Q9418","display_name":"Psychology","level":0,"score":0.4666000008583069},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.4620000123977661},{"id":"https://openalex.org/C204321447","wikidata":"https://www.wikidata.org/wiki/Q30642","display_name":"Natural language processing","level":1,"score":0.4169999957084656},{"id":"https://openalex.org/C77265313","wikidata":"https://www.wikidata.org/wiki/Q879844","display_name":"Rest (music)","level":2,"score":0.3804999887943268},{"id":"https://openalex.org/C2778334786","wikidata":"https://www.wikidata.org/wiki/Q1586270","display_name":"Variation (astronomy)","level":2,"score":0.3695000112056732},{"id":"https://openalex.org/C168167062","wikidata":"https://www.wikidata.org/wiki/Q1117970","display_name":"Component (thermodynamics)","level":2,"score":0.362199991941452},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.34459999203681946},{"id":"https://openalex.org/C155512373","wikidata":"https://www.wikidata.org/wiki/Q287450","display_name":"Residual","level":2,"score":0.3375000059604645},{"id":"https://openalex.org/C2777211547","wikidata":"https://www.wikidata.org/wiki/Q17141490","display_name":"Training (meteorology)","level":2,"score":0.30799999833106995},{"id":"https://openalex.org/C2987567764","wikidata":"https://www.wikidata.org/wiki/Q125421","display_name":"Second language","level":2,"score":0.3052000105381012},{"id":"https://openalex.org/C51632099","wikidata":"https://www.wikidata.org/wiki/Q3985153","display_name":"Training set","level":2,"score":0.2867000102996826},{"id":"https://openalex.org/C81917197","wikidata":"https://www.wikidata.org/wiki/Q628760","display_name":"Selection (genetic algorithm)","level":2,"score":0.2849000096321106},{"id":"https://openalex.org/C151201525","wikidata":"https://www.wikidata.org/wiki/Q177239","display_name":"Limit (mathematics)","level":2,"score":0.2815000116825104},{"id":"https://openalex.org/C2778883600","wikidata":"https://www.wikidata.org/wiki/Q2390977","display_name":"Language proficiency","level":2,"score":0.2773999869823456},{"id":"https://openalex.org/C2778571376","wikidata":"https://www.wikidata.org/wiki/Q1355821","display_name":"Frontier","level":2,"score":0.2667999863624573},{"id":"https://openalex.org/C42058472","wikidata":"https://www.wikidata.org/wiki/Q810214","display_name":"Base (topology)","level":2,"score":0.26080000400543213},{"id":"https://openalex.org/C195324797","wikidata":"https://www.wikidata.org/wiki/Q33742","display_name":"Natural language","level":2,"score":0.25929999351501465},{"id":"https://openalex.org/C4554734","wikidata":"https://www.wikidata.org/wiki/Q593744","display_name":"Knowledge base","level":2,"score":0.2565000057220459}],"mesh":[],"locations_count":2,"locations":[{"id":"pmh:oai:arXiv.org:2608.21766","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2608.21766","pdf_url":"https://arxiv.org/pdf/2608.21766","source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"host_organization_lineage_names":[],"type":"repository"},"license":null,"license_id":null,"version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},{"id":"doi:10.48550/arxiv.2608.21766","doi":"https://doi.org/10.48550/arxiv.2608.21766","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2608.21766","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"host_organization_lineage_names":[],"type":"repository"},"license":null,"license_id":null,"version":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Preprint"}],"best_oa_location":{"id":"pmh:oai:arXiv.org:2608.21766","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2608.21766","pdf_url":"https://arxiv.org/pdf/2608.21766","source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"host_organization_lineage_names":[],"type":"repository"},"license":null,"license_id":null,"version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},"sustainable_development_goals":[],"study_designs":[],"awards":[],"funders":[],"has_content":{"pdf":true,"grobid_xml":true},"content_urls":{"pdf":"https://content.openalex.org/works/W7204185906.pdf","grobid_xml":"https://content.openalex.org/works/W7204185906.grobid-xml"},"referenced_works_count":23,"referenced_works":["https://openalex.org/W2529194139","https://openalex.org/W3201174429","https://openalex.org/W3210904070","https://openalex.org/W4294435850","https://openalex.org/W4386081218","https://openalex.org/W4386494224","https://openalex.org/W4390833079","https://openalex.org/W4405173927","https://openalex.org/W4414697811","https://openalex.org/W4415314106","https://openalex.org/W4416085925","https://openalex.org/W4416100856","https://openalex.org/W4416258840","https://openalex.org/W4417114549","https://openalex.org/W7094947999","https://openalex.org/W7131428749","https://openalex.org/W7140091091","https://openalex.org/W7155039456","https://openalex.org/W7160654208","https://openalex.org/W7162339950","https://openalex.org/W7162628876","https://openalex.org/W7165638749","https://openalex.org/W7197013104"],"related_works":["https://openalex.org/W4417114549","https://openalex.org/W4416100856","https://openalex.org/W7162339950","https://openalex.org/W7159916509","https://openalex.org/W7207546818","https://openalex.org/W7165638749","https://openalex.org/W7094947999","https://openalex.org/W7166696366","https://openalex.org/W7214016586","https://openalex.org/W4416258840"],"abstract_inverted_index":{"Both":[0],"capability":[1],"and":[2,37,54,78,81,114,169,183,221,258],"safety":[3],"benchmarks":[4],"rest":[5],"upon":[6],"the":[7,10,97,122,147,193,200,216,225,241,248],"assumption":[8,26],"that":[9,32,140,204,229],"behavior":[11,22],"of":[12,64,150,218,227],"language":[13,56,73],"models":[14,30,57,74,252],"undergoing":[15],"a":[16,61,197],"test":[17,128],"is":[18,93,103,143,207],"informative":[19],"about":[20],"their":[21,39,106,119,167,259],"in":[23,52,105,163,174],"deployment.":[24],"This":[25,44],"can":[27,191],"fail,":[28],"should":[29],"infer":[31],"they":[33,256],"are":[34,172],"being":[35,90],"evaluated":[36],"condition":[38],"response":[40],"on":[41],"such":[42],"context.":[43],"hypothesis,":[45],"termed":[46],"``evaluation":[47],"awareness'',":[48],"has":[49],"been":[50],"observed":[51],"frontier":[53],"open-weight":[55],"alike.":[58],"We":[59],"provide":[60],"systematic":[62],"study":[63],"this":[65],"phenomenon,":[66],"by":[67,111],"probing":[68],"for":[69,243,247],"it":[70,102],"across":[71,180,199],"six":[72],"(from":[75],"four":[76],"families":[77],"three":[79,82],"sizes)":[80],"metrics.":[83],"More":[84],"precisely,":[85],"we":[86,126,138],"examine":[87],"whether":[88],"(i)":[89],"under":[91],"evaluation":[92,141,205],"linearly":[94,144],"represented":[95],"within":[96,210],"models'":[98],"activations":[99],"space,":[100],"(ii)":[101],"verbalized":[104],"output":[107],"tokens":[108],"(as":[109],"scored":[110],"an":[112],"LLM-as-judge),":[113],"(iii)":[115],"steering":[116,187],"causally":[117],"affects":[118],"behavior.":[120],"For":[121],"open-checkpoint":[123],"Olmo":[124,201],"models,":[125,181,212],"further":[127],"these":[129,159],"measures":[130],"at":[131,233],"every":[132,151,234],"training":[133,236],"stage.":[134,237],"In":[135],"doing":[136],"so,":[137],"report":[139],"awareness":[142,206],"decodable":[145],"from":[146],"residual":[148],"streams":[149],"model":[152],"(best":[153],"AUROC":[154],"$\\geq":[155],"0.7$).":[156],"By":[157],"contrast,":[158],"representations":[160],"align":[161],"only":[162],"part":[164],"with":[165],"verbalization:":[166],"correlations":[168],"mutual":[170],"information":[171],"nonzero":[173],"some":[175],"settings,":[176],"yet":[177],"vary":[178],"substantially":[179],"layers,":[182],"readout":[184],"choices.":[185],"Nevertheless,":[186],"along":[188],"probe-derived":[189],"directions":[190],"shift":[192],"verbalization":[194],"scores.":[195],"Finally,":[196],"comparison":[198],"checkpoints":[202],"reveals":[203],"already":[208],"present":[209],"base":[211],"becomes":[213],"amplified":[214],"throughout":[215],"stages":[217],"supervised":[219],"fine-tuning,":[220],"remains":[222],"stable":[223],"thereafter---unlike":[224],"effects":[226],"steering,":[228],"grow":[230],"more":[231],"pronounced":[232],"successive":[235],"These":[238],"results":[239],"show":[240],"need":[242],"evaluations":[244],"to":[245],"account":[246],"disjunction":[249],"between":[250],"what":[251,255],"represent":[253],"internally,":[254],"verbalize,":[257],"steering.":[260]},"counts_by_year":[{"year":2026,"cited_by_count":3}],"updated_date":"2026-11-07T11:25:07.637180","created_date":"2026-08-26T00:00:00"}