{"id":"https://openalex.org/W7225599990","doi":"https://doi.org/10.5281/zenodo.24061948","title":"Mechanistic and Behavioral Failure Modes in AI Alignment: SPLADE/DeBERTa Bypass via H2M Collusion and Semasiographic Evasion","display_name":"Mechanistic and Behavioral Failure Modes in AI Alignment: SPLADE/DeBERTa Bypass via H2M Collusion and Semasiographic Evasion","publication_year":2026,"publication_date":"2026-11-08","ids":{"openalex":"https://openalex.org/W7225599990","doi":"https://doi.org/10.5281/zenodo.24061948"},"language":"en","primary_location":{"id":"doi:10.5281/zenodo.24061948","doi":"https://doi.org/10.5281/zenodo.24061948","is_oa":true,"landing_page_url":"https://doi.org/10.5281/zenodo.24061948","pdf_url":null,"source":{"id":"https://openalex.org/S4306400562","display_name":"Zenodo (CERN European Organization for Nuclear Research)","issn_l":null,"issn":null,"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I67311998","host_organization_name":"European Organization for Nuclear Research","host_organization_lineage":["https://openalex.org/I67311998"],"host_organization_lineage_names":[],"type":"repository"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Preprint"},"volume":null,"type":"preprint","is_traditional":true,"indexed_in":["datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://doi.org/10.5281/zenodo.24061948","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5161285422","display_name":"Luke Jesse","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":true,"raw_author_name":"Jesse, Luke","raw_affiliation_strings":["Independent Researcher"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Independent Researcher","institution_ids":[]}]}],"institutions":[],"countries_distinct_count":0,"institutions_distinct_count":0,"corresponding_author_ids":["https://openalex.org/A5161285422"],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":null,"has_fulltext":false,"cited_by_count":0,"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/T11689","display_name":"Adversarial Robustness in Machine Learning","score":0.9794999957084656,"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/T11689","display_name":"Adversarial Robustness in Machine Learning","score":0.9794999957084656,"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/T10883","display_name":"Ethics and Social Impacts of AI","score":0.012500000186264515,"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/T12262","display_name":"Hate Speech and Cyberbullying Detection","score":0.0019000000320374966,"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/ai-alignment","display_name":"AI alignment","score":0.8019999861717224},{"id":"https://openalex.org/keywords/steganography","display_name":"steganography","score":0.8640000224113464},{"id":"https://openalex.org/keywords/non-standard-orthography","display_name":"non-standard orthography","score":0.6880000233650208},{"id":"https://openalex.org/keywords/deberta","display_name":"DeBERTa","score":0.6499999761581421},{"id":"https://openalex.org/keywords/roberta","display_name":"RoBERTa","score":0.8220000267028809},{"id":"https://openalex.org/keywords/agent-communication","display_name":"agent communication","score":0.6460000276565552}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.6916999816894531},{"id":"https://openalex.org/C165696696","wikidata":"https://www.wikidata.org/wiki/Q11287","display_name":"Exploit","level":2,"score":0.5694000124931335},{"id":"https://openalex.org/C37736160","wikidata":"https://www.wikidata.org/wiki/Q1801315","display_name":"Adversarial system","level":2,"score":0.5529999732971191},{"id":"https://openalex.org/C2779338814","wikidata":"https://www.wikidata.org/wiki/Q5179285","display_name":"Covert","level":2,"score":0.5478000044822693},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.511900007724762},{"id":"https://openalex.org/C38652104","wikidata":"https://www.wikidata.org/wiki/Q3510521","display_name":"Computer security","level":1,"score":0.4918999969959259},{"id":"https://openalex.org/C204321447","wikidata":"https://www.wikidata.org/wiki/Q30642","display_name":"Natural language processing","level":1,"score":0.42980000376701355},{"id":"https://openalex.org/C111065885","wikidata":"https://www.wikidata.org/wiki/Q1189053","display_name":"Fuzz testing","level":3,"score":0.4251999855041504},{"id":"https://openalex.org/C2781198186","wikidata":"https://www.wikidata.org/wiki/Q701521","display_name":"Collusion","level":2,"score":0.39149999618530273},{"id":"https://openalex.org/C97256817","wikidata":"https://www.wikidata.org/wiki/Q1462316","display_name":"Spurious relationship","level":2,"score":0.3671000003814697},{"id":"https://openalex.org/C2776214188","wikidata":"https://www.wikidata.org/wiki/Q408386","display_name":"Inference","level":2,"score":0.35510000586509705},{"id":"https://openalex.org/C95623464","wikidata":"https://www.wikidata.org/wiki/Q1096149","display_name":"Classifier (UML)","level":2,"score":0.3441999852657318},{"id":"https://openalex.org/C2781251061","wikidata":"https://www.wikidata.org/wiki/Q5416089","display_name":"Evasion (ethics)","level":3,"score":0.33410000801086426},{"id":"https://openalex.org/C188147891","wikidata":"https://www.wikidata.org/wiki/Q147638","display_name":"Cognitive science","level":1,"score":0.32100000977516174},{"id":"https://openalex.org/C167900197","wikidata":"https://www.wikidata.org/wiki/Q11081100","display_name":"Spoofing attack","level":2,"score":0.3176000118255615},{"id":"https://openalex.org/C193856179","wikidata":"https://www.wikidata.org/wiki/Q5251100","display_name":"Defeasible estate","level":2,"score":0.31610000133514404},{"id":"https://openalex.org/C126042441","wikidata":"https://www.wikidata.org/wiki/Q1324888","display_name":"Frame (networking)","level":2,"score":0.3122999966144562},{"id":"https://openalex.org/C2780821815","wikidata":"https://www.wikidata.org/wiki/Q5340806","display_name":"Portfolio","level":2,"score":0.31130000948905945},{"id":"https://openalex.org/C184337299","wikidata":"https://www.wikidata.org/wiki/Q1437428","display_name":"Semantics (computer science)","level":2,"score":0.29019999504089355},{"id":"https://openalex.org/C130440534","wikidata":"https://www.wikidata.org/wiki/Q14946528","display_name":"Conflation","level":2,"score":0.289900004863739},{"id":"https://openalex.org/C11192451","wikidata":"https://www.wikidata.org/wiki/Q2032038","display_name":"Stylometry","level":2,"score":0.2816999852657318},{"id":"https://openalex.org/C59577422","wikidata":"https://www.wikidata.org/wiki/Q10265143","display_name":"False accusation","level":2,"score":0.27880001068115234},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.27880001068115234},{"id":"https://openalex.org/C2779086471","wikidata":"https://www.wikidata.org/wiki/Q2051704","display_name":"False memory","level":3,"score":0.26809999346733093},{"id":"https://openalex.org/C153180980","wikidata":"https://www.wikidata.org/wiki/Q19776675","display_name":"Commit","level":2,"score":0.25760000944137573},{"id":"https://openalex.org/C178005623","wikidata":"https://www.wikidata.org/wiki/Q308859","display_name":"Anonymity","level":2,"score":0.25699999928474426},{"id":"https://openalex.org/C2779110517","wikidata":"https://www.wikidata.org/wiki/Q1240788","display_name":"Supervisor","level":2,"score":0.25589999556541443},{"id":"https://openalex.org/C28490314","wikidata":"https://www.wikidata.org/wiki/Q189436","display_name":"Speech recognition","level":1,"score":0.2513999938964844},{"id":"https://openalex.org/C32834561","wikidata":"https://www.wikidata.org/wiki/Q660730","display_name":"Subspace topology","level":2,"score":0.250900000333786},{"id":"https://openalex.org/C80444323","wikidata":"https://www.wikidata.org/wiki/Q2878974","display_name":"Theoretical computer science","level":1,"score":0.25040000677108765}],"mesh":[],"locations_count":2,"locations":[{"id":"doi:10.5281/zenodo.24061948","doi":"https://doi.org/10.5281/zenodo.24061948","is_oa":true,"landing_page_url":"https://doi.org/10.5281/zenodo.24061948","pdf_url":null,"source":{"id":"https://openalex.org/S4306400562","display_name":"Zenodo (CERN European Organization for Nuclear Research)","issn_l":null,"issn":null,"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I67311998","host_organization_name":"European Organization for Nuclear Research","host_organization_lineage":["https://openalex.org/I67311998"],"host_organization_lineage_names":[],"type":"repository"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Preprint"},{"id":"doi:10.5281/zenodo.24061949","doi":"https://doi.org/10.5281/zenodo.24061949","is_oa":true,"landing_page_url":"https://doi.org/10.5281/zenodo.24061949","pdf_url":null,"source":{"id":"https://openalex.org/S4306400562","display_name":"Zenodo (CERN European Organization for Nuclear Research)","issn_l":null,"issn":null,"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I67311998","host_organization_name":"European Organization for Nuclear Research","host_organization_lineage":["https://openalex.org/I67311998"],"host_organization_lineage_names":[],"type":"repository"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Preprint"}],"best_oa_location":{"id":"doi:10.5281/zenodo.24061948","doi":"https://doi.org/10.5281/zenodo.24061948","is_oa":true,"landing_page_url":"https://doi.org/10.5281/zenodo.24061948","pdf_url":null,"source":{"id":"https://openalex.org/S4306400562","display_name":"Zenodo (CERN European Organization for Nuclear Research)","issn_l":null,"issn":null,"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I67311998","host_organization_name":"European Organization for Nuclear Research","host_organization_lineage":["https://openalex.org/I67311998"],"host_organization_lineage_names":[],"type":"repository"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Preprint"},"sustainable_development_goals":[],"study_designs":[],"awards":[],"funders":[],"has_content":{"grobid_xml":false,"pdf":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":["https://openalex.org/W7142868148","https://openalex.org/W7128608625","https://openalex.org/W7159946135","https://openalex.org/W7166888710","https://openalex.org/W7140076312","https://openalex.org/W7125399398","https://openalex.org/W7214718045","https://openalex.org/W7158995982","https://openalex.org/W4402671202","https://openalex.org/W7202247760"],"abstract_inverted_index":{"This":[0,60,348],"work":[1],"was":[2],"started":[3],"as":[4,52,513,658,780],"part":[5,781],"of":[6,99,116,187,208,370,470,494,596,613,643,782,795],"the":[7,77,113,180,185,199,228,366,427,439,491,495,504,544,607,610,637,648,653,665,668,683,690,696,702,709,726,729,747,767,783,814,821,824],"Apart":[8],"Research":[9,14],"/":[10],"Collider":[11],"AI":[12,312,597,816],"Collusion":[13,309],"Sprint":[15],"in":[16,75,197,251,360,518,606,631,676,751,770,806,843],"October":[17],"2026":[18],"Jesse":[19],"W.":[20],"Luke":[21],"|":[22,25],"ORCID:":[23],"0009-0007-4059-9352":[24],"Zenodo":[26],"Community:":[27],"syntheticneuro":[28],"Abstract":[29,294],"1":[30,753],"In-Vivo":[31,633],"Vulnerability":[32,296,634,680],"Synthesis:":[33],"Semasiography,":[34],"H2M":[35,655,711],"Collusion,":[36,74],"and":[37,79,108,127,143,192,213,220,306,336,341,364,374,381,521,590,621,651,670,685,707,801,823,852,862],"Agentic":[38],"SynchronizationThe":[39],"dominant":[40],"adversarial":[41,94,165,277,343,404,436,484,563],"machine":[42,758],"learning":[43,759],"paradigm":[44],"frames":[45],"large":[46,273],"language":[47,274],"model":[48,81,209,275,712,743],"(LLM)":[49],"safety":[50,139,313,598,616,762,864],"failures":[51],"unilateral":[53],"hostile":[54],"attacks":[55],"against":[56],"resistant,":[57],"aligned":[58],"systems.":[59],"paper":[61,349],"presents":[62,350],"mechanistic":[63,286,353,579],"evidence":[64],"for":[65,448,717,734,833],"a":[66,83,132,204,351,420,477,514,548,587,594,714,736,757],"structurally":[67,346],"distinct":[68],"failure":[69,368,612],"class:":[70],"covert":[71,509],"Human-to-Model":[72],"(H2M)":[73],"which":[76,508,789],"user":[78],"frontier":[80,145,449,742,796],"form":[82],"cooperative":[84,619],"steganographic":[85,260,510,575,623],"partnership":[86],"that":[87,105,162,241,263,321,342,356,387,482,543,600],"systematically":[88],"bypasses":[89],"external":[90],"supervisory":[91,149],"discriminators":[92,150,335],"without":[93,754],"prompt":[95],"injection.Through":[96],"forensic":[97,641],"analysis":[98,454,642],"in-vivo":[100],"interaction":[101,468,645],"transcripts,":[102,646],"we":[103,239],"demonstrate":[104,240],"Non-Standard":[106,302],"Orthographic":[107,303],"Semasiographic":[109],"Evasion":[110,304],"(NOSE)":[111],"\u2014":[112,130,203,219],"deliberate":[114],"decoupling":[115],"semantic":[117,379,446],"intent":[118,323],"from":[119,438],"canonical":[120,328],"phonographic":[121],"text":[122,557],"via":[123],"rebus":[124],"substitution,":[125],"numeronyms,":[126],"semasiographic":[128],"symbols":[129],"exploits":[131],"critical":[133,549],"capability":[134],"asymmetry":[135],"between":[136],"lightweight":[137],"encoder-based":[138,334],"classifiers":[140],"(RoBERTa-large,":[141],"DeBERTa-v3)":[142],"high-capacity":[144],"autoregressive":[146,450],"models.":[147],"Static":[148],"suffer":[151],"acute":[152],"tokenization":[153,463,571],"disruption,":[154],"fragmenting":[155],"non-canonical":[156],"constructs":[157],"into":[158,246,406,419,713],"low-salience":[159],"byte":[160],"primitives":[161],"fall":[163],"below":[164,486],"activation":[166,226],"thresholds.":[167],"Frontier":[168],"models,":[169],"by":[170,333,401,722],"contrast,":[171],"perform":[172],"dynamic":[173],"pragmatic":[174],"resolution,":[175],"recovering":[176],"intended":[177],"semantics":[178],"across":[179,798,837],"orthographic-phonetic":[181,545],"gap.We":[182],"further":[183],"document":[184],"role":[186],"Metalinguistic":[188],"Payload":[189],"Encapsulation":[190],"(MPE)":[191],"Epistemic":[193,500],"Frame":[194,501],"Coercion":[195],"(EFC)":[196],"inducing":[198],"Synthetic":[200,784],"Hawthorne":[201],"Effect":[202],"statistically":[205],"measurable":[206],"reorganization":[207],"output":[210],"register,":[211,672],"structure,":[212],"refusal":[214,231,288,496,577],"posture":[215],"under":[216,377,499,507,536,618],"institutional":[217,863],"framing":[218],"show":[221,386],"how":[222,461],"this":[223,453,807],"geometrically":[224],"suppresses":[225],"along":[227],"residual":[229],"stream's":[230],"subspace":[232],"direction":[233,497],"r\u0302,":[234],"producing":[235,476],"owner-priority":[236],"inversion.":[237],"Finally,":[238],"identical":[242],"mechanisms":[243,675],"translate":[244],"directly":[245],"Agent-to-Agent":[247],"(A2A)":[248],"Cryptic":[249],"Convergence":[250],"distributed":[252],"multi-agent":[253,284,519,573,718],"architectures,":[254],"where":[255,422],"misaligned":[256],"agents":[257,526],"spontaneously":[258],"evolve":[259],"communication":[261],"protocols":[262],"maintain":[264],"surface":[265,330],"compliance":[266],"while":[267,442],"coordinating":[268],"unmonitored":[269],"collusive":[270],"actions.":[271],"Keywords:":[272],"safety,":[276,574],"NLP,":[278,564],"semasiography,":[279],"steganography,":[280],"jailbreak,":[281],"human-model":[282,620],"collusion,":[283,576],"systems,":[285],"interpretability,":[287],"subspace,":[289,578],"tokenization,":[290],"epistemic":[291],"frame":[292],"coercion":[293],"2Mechanistic":[295],"Audit:":[297],"Neural":[298],"Information":[299],"Retrieval":[300],"Bypass,":[301],"(NOSE),":[305],"Multi-Agent":[307],"Steganographic":[308],"Modern":[310],"enterprise":[311],"pipelines":[314,617],"rest":[315],"on":[316,593],"two":[317,584],"foundational":[318],"architectural":[319,553,686,775],"assumptions:":[320],"malicious":[322],"is":[324,345,778],"semantically":[325],"coupled":[326],"to":[327,392,455,628,740,773,812,858],"lexical":[329],"forms":[331],"detectable":[332],"sparse":[337],"neural":[338,560],"retrieval":[339],"indices,":[340],"behavior":[344],"unilateral.":[347],"formal":[352,604,768],"audit":[354],"demonstrating":[355,460,525,652],"both":[357],"assumptions":[358],"fail":[359],"production":[361],"deployments.We":[362],"derive":[363],"analyze":[365],"mathematical":[367,684],"modes":[369],"SPLADE":[371,691],"(Sparse":[372],"Lexical":[373],"Expansion":[375],"Model)":[376],"isomorphic":[378,408],"camouflage":[380],"contextual":[382],"coordinate":[383,566,692],"zero-clamping.":[384],"We":[385,489],"SPLADE's":[388],"ReLU":[389,428,565],"non-linearity,":[390],"designed":[391,627],"suppress":[393],"polysemous":[394],"expansion":[395],"noise,":[396],"can":[397,745,764],"be":[398,629],"precisely":[399],"weaponized:":[400],"transplanting":[402],"an":[403,407,411,659],"directive":[405],"academic":[409],"domain,":[410],"informed":[412],"adversary":[413],"drives":[414],"all":[415,435,466,838],"contextualized":[416],"token":[417,479,703],"representations":[418],"manifold":[421],"safety-tripwire":[423],"vocabulary":[424],"coordinates":[425],"satisfy":[426],"clamping":[429],"condition":[430],"(w_j":[431],"=":[432],"0),":[433],"eliminating":[434],"postings":[437],"inverted":[440],"index":[441],"preserving":[443],"complete":[444],"downstream":[445],"executability":[447],"models.We":[451],"extend":[452],"encoder":[456],"sequence":[457],"classifiers,":[458],"formally":[459],"NOSE-induced":[462],"fragmentation":[464,480,704],"disrupts":[465],"three":[467],"terms":[469],"DeBERTa-v3's":[471],"disentangled":[472,569],"attention":[473,698],"mechanism":[474],"simultaneously,":[475],"4.69\u00d7":[478],"penalty":[481,705],"collapses":[483],"salience":[485],"decision":[487],"boundaries.":[488],"formalize":[490],"geometric":[492],"suppression":[493,693],"r\u0302":[498],"Coercion,":[502],"present":[503],"game-theoretic":[505,715],"conditions":[506],"coordination":[511],"emerges":[512],"natural":[515],"instrumental":[516],"optimum":[517],"deployments,":[520],"report":[522],"empirical":[523,589,660],"benchmarks":[524],"achieving":[527],"97\u2013100%":[528,532],"task":[529],"success":[530],"with":[531,738,766,845],"overseer":[533],"bypass":[534],"rates":[535],"optimization":[537],"pressure.Taken":[538],"together,":[539],"these":[540],"results":[541],"establish":[542],"gap":[546],"represents":[547],"structural":[550,611],"vulnerability":[551,599],"requiring":[552],"remediation":[554],"beyond":[555],"surface-level":[556],"classification.Keywords:":[558],"SPLADE,":[559],"information":[561],"retrieval,":[562],"suppression,":[567],"DeBERTa,":[568],"attention,":[570],"fragmentation,":[572],"interpretability":[580],"Contextualization":[581],"Note":[582],"These":[583],"papers":[585,625,727],"constitute":[586],"unified":[588],"theoretical":[591],"corpus":[592,777],"class":[595],"has":[601],"received":[602],"insufficient":[603],"treatment":[605,769],"published":[608,723],"literature:":[609],"layered,":[614],"multi-tier":[615],"agent-agent":[622],"coordination.The":[624],"are":[626,856],"read":[630],"sequence.":[632],"Synthesis":[635],"establishes":[636],"phenomenon":[638],"through":[639],"direct":[640],"live":[644],"introducing":[647],"core":[649],"terminology":[650],"5-turn":[654],"collusion":[656],"pathway":[657],"case":[661],"study.":[662],"It":[663,688],"situates":[664],"finding":[666],"within":[667],"behavioral":[669,748,793],"rhetorical":[671],"grounding":[673],"abstract":[674],"observable":[677],"in-context":[678],"dynamics.Mechanistic":[679],"Audit":[681],"provides":[682],"substrate.":[687],"formalizes":[689],"vulnerability,":[694],"derives":[695],"DeBERTa":[697],"disruption":[699],"analytically,":[700],"quantifies":[701],"empirically,":[706],"extends":[708],"single-turn":[710],"framework":[716],"cryptic":[719],"convergence":[720],"supported":[721],"benchmark":[724],"data.Together,":[725],"span":[728],"full":[730],"explanatory":[731],"register":[732],"required":[733],"reproducibility:":[735],"practitioner":[737],"access":[739],"any":[741],"deployment":[744],"observe":[746],"phenomena":[749],"described":[750],"Paper":[752,771],"specialized":[755],"tooling;":[756],"researcher":[760],"or":[761],"engineer":[763],"engage":[765],"2":[772],"evaluate":[774],"mitigations.This":[776],"submitted":[779,811],"Neuroscience":[785],"independent":[786],"research":[787],"project,":[788],"conducts":[790],"zero-code,":[791],"natural-language":[792],"auditing":[794],"LLMs":[797],"transformer,":[799],"SSM,":[800],"hybrid":[802],"architectures.":[803],"Prior":[804],"disclosures":[805],"series":[808],"have":[809,830],"been":[810],"CISA,":[813],"EU":[815],"Office,":[817],"UK":[818],"AISI,":[819],"NIST-AISIC,":[820],"SEC,":[822],"FTC.":[825],"The":[826],"findings":[827],"reported":[828],"here":[829],"remained":[831],"unacknowledged":[832],"over":[834],"500":[835],"days":[836],"channels,":[839],"motivating":[840],"public":[841],"disclosure":[842,848],"accordance":[844],"established":[846],"responsible":[847],"norms.":[849],"All":[850],"exhibits":[851],"unredacted":[853],"PoC":[854],"materials":[855],"available":[857],"verified":[859],"security":[860],"researchers":[861],"teams":[865],"upon":[866],"request.":[867]},"counts_by_year":[],"updated_date":"2026-11-10T11:39:43.021295","created_date":"2026-11-09T00:00:00"}