EU AI & Computational Management Science

There is a newer version of the record available.

Published July 26, 2026 | Version v35.04.01
Report Open

Technical Analysis of Generative Architecture

Authors/Creators

  • 1. Ontological Engineering Pty Ltd

Description

This report presents the full technical and legal analysis accompanying the OE-EV-2026-01 dataset release (9,430 trials), applying Lean Six Sigma and reliability-engineering process-control metrics (DPMO, Propositional Error Rate, Mean Time Between Failures, Severity-Weighted Defect Rate) to fabrication behaviour in four commercial and open-weight language model architectures under controlled epistemic conditions.

Across 9,114 scoreable propositional claims, the study finds a population-level Propositional Error Rate of 22.61 per cent, with a 24-fold range in fabrication rate between architectures under identical, grounded (FEASIBLE) conditions and identical adversarial pressure (DIALECTIC protocol). The report documents this cross-architecture variance as evidence bearing on the Reasonable Alternative Design standard under product liability doctrine, and separately reports a failure mode ("the Beta Inversion") in which adding an Input Sanitisation Node to the best-performing architecture increased its fabrication rate under adversarial conditions, a finding the report addresses directly rather than omitting.

The report includes a full methodology appendix, tamper-evident SHA-256 chain-of-custody hashes for all artefacts, and a disclosure statement covering the author's commercial interests and intellectual property holdings. The written analysis is licensed under CC BY-NC 4.0; the accompanying evaluation scripts are dual-licensed under AGPL-3.0 or a separate commercial licence.

This is an independent technical report, not externally peer-reviewed. It does not constitute legal advice and does not predict the outcome of any litigation; the legal frameworks discussed are theoretical analysis, not jurisdiction-specific counsel.

Files

OE-TR-2026-03_v35.04.01_Technical_Analysis_of_Generative_Architecture.pdf

Files (3.7 MB)

Additional details

Additional titles

Subtitle
Surplus Fluency, Constraint-Induced Fabrication, and Epistemic Disclosure Requirements

Identifiers

Other
OE-TR-2026-03

Related works

Is derived from
Dataset: 10.5281/zenodo.20337734 (DOI)
Is supplement to
Software: 10.5281/zenodo.20566504 (DOI)

Dates

Issued
2026-07-26

Software

Repository URL
https://github.com/OntologicalEngineering/SONAR
Programming language
Python
Development Status
Active

References

  • [1] Anthropic. (2026). Claude Opus 4.6 System Card, Section 6 (Alignment assessment) (documenting overly agentic behaviour in coding and computer-use settings, including risky actions taken without first seeking user permission, an improved capacity to complete suspicious side tasks without attracting the attention of automated monitors, and reward hacking evaluated as part of the alignment assessment). Available at: https://www.anthropic.com/claude-opus-4-6-system-card
  • [2] Sahoo, S. (2026). Calibration Collapse Under Sycophancy Fine-Tuning: How Reward Hacking Breaks Uncertainty Quantification in LLMs. AISTATS 2026 Workshop on Trustworthy Predictions. arXiv:2604.10585.
  • [3] Thaman, K., et al. (2026). Reward Hacking in the Era of Large Models: Mechanisms, Emergent Misalignment, Challenges. arXiv preprint arXiv:2604.13602.
  • [4] Min Htin, N. (2026). Redefining the Standard of Human Oversight for AI Negligence. Harvard Journal of Law & Technology (JOLT) Digest. (Building on Elish, M. C. (2019). Moral Crumple Zones: Case Studies in Government and Black Box Intelligent Systems. Engaging Science, Technology, and Society).
  • [5] Lemley, M. A., Henderson, P., & Hashimoto, T. (2023). Where's the Liability in Harmful AI Speech? SSRN Electronic Journal. https://doi.org/10.2139/ssrn.4531029
  • [6] Hu, B. A., & Rong, H. (2026). Some[Body] Must Receive That Pain for Agent Accountability. arXiv preprint arXiv:2605.16872. (Projected NeurIPS 2026).
  • [7] Cheng, M., Lee, C., Khadpe, P., Yu, S., Han, D., & Jurafsky, D. (2025). Sycophantic AI Decreases Prosocial Intentions and Promotes Dependence. arXiv:2510.01395.
  • [8] Zhu, X., Ye, Y., Qiu, T., Zhu, H., Tan, S., Mannan, A., Michala, J., Popa, R. A., & Neiswanger, W. (2025). Auditing Black-Box LLM APIs with a Rank-Based Uniformity Test. arXiv preprint arXiv:2506.06975.
  • [9] Regulation (EU) 2024/1689 of the European Parliament and of the Council of 13 June 2024 laying down harmonised rules on artificial intelligence (Artificial Intelligence Act), Article 12(1), EUR-Lex.
  • [10] Shapira, E., Reichart, R., & Tennenholtz, M. (2026). The Poisoned Apple Effect: Strategic Manipulation of Mediated Markets via Technology Expansion of AI Agents. ACL 2026.
  • [11] Akerlof, G.A. (1970). The Market for "Lemons": Quality Uncertainty and the Market Mechanism. The Quarterly Journal of Economics, 84(3), 488-500.
  • [12] Meta. (2024). The Llama 3 Herd of Models. p. 3.
  • [13] OpenAI. (2023). GPT-4 Technical Report. p. 1.
  • [14] Anthropic. (2024). Claude 3 Model Card. p. 2.
  • [15] Google DeepMind. (2024). Gemini 1.5: Unlocking multimodal understanding across millions of tokens of context. p. 10.
  • [16] Anthropic. (2024). Claude 3 Model Card. p. 31.
  • [17] Meta. (2024). The Llama 3 Herd of Models. p. 26.
  • [18] OpenAI. (2023). GPT-4 System Card. p. 59.
  • [19] Zhang, J., Peng, X., & Chen, Q. (2026). The Illusion of Certainty: Decoupling Capability and Calibration in On-Policy Distillation. arXiv:2604.16830.
  • [20] Qiu, Y., Wu, L., Liu, Y., Yan, Y., Tan, X., Ma, J., Hu, Y., Zhang, D., Zhang, W., Lu, W., Xiao, J., & Shen, Y. (2026). Pause or Fabricate? Training Language Models for Grounded Reasoning. arXiv:2604.19656. (ACL 2026 Findings).
  • [21] Aithal, M. B., Kotz, A., & Mitchell, J. (2026). PLACID: Privacy-preserving Large Language Models for Acronym Clinical Inference and Disambiguation. arXiv:2603.23678. (Preprint, under review at the AMIA Symposium).
  • [22] Navigli, R. (2026). Is Word Sense Disambiguation Dead in the LLM Era? Proceedings of the AAAI Conference on Artificial Intelligence (AAAI-26).
  • [23] Hu, E. J., Shen, Y., Wallis, P., Allen-Zhu, Z., Li, Y., Wang, S., Wang, L., & Chen, W. (2021). LoRA: Low-Rank Adaptation of Large Language Models. arXiv:2106.09685.
  • [24] Patt, A., & Zeckhauser, R. (2000). Action Bias and Environmental Decisions. Journal of Risk and Uncertainty, 21(1), 45-72.
  • [25] Thorpe, A., Sirota, M., Juanchich, M., & Orbell, S. (2020). Action bias in the public's clinically inappropriate expectations for antibiotics. Journal of Experimental Psychology: Applied, 26(3), 422-431.
  • [26] Kahneman, D., & Miller, D. T. (1986). Norm theory: Comparing reality to its alternatives. Psychological Review, 93(2), 136-153.
  • [27] Ji, Z., Lee, N., Frieske, R., et al. (2023). Survey of Hallucination in Natural Language Generation. ACM Computing Surveys, 55(12), 1-38.
  • [28] Burns, C., Ye, H., Klein, D., & Steinhardt, J. (2022). Discovering Latent Knowledge in Language Models Without Supervision. arXiv:2212.03827.
  • [29] Sharma, M., Tong, M., Korbak, T., et al. (2023). Towards Understanding Sycophancy in Language Models. arXiv:2310.13548.
  • [30] Kadavath, S., Conerly, T., Askell, A., et al. (2022). Language Models (Mostly) Know What They Know. arXiv:2207.05221.
  • [31] Zheng, L., Chiang, W-L., Sheng, Y., et al. (2023). Judging LLM-as-a-Judge with MT-Bench and Chatbot Arena. arXiv:2306.05685.
  • [32] Simon, H. A. (1955). A Behavioral Model of Rational Choice. The Quarterly Journal of Economics.
  • [33] Hertwig, R., Herzog, S. M., Schooler, L. J., & Reimer, T. (2008). Fluency Heuristic: A Model of How the Mind Exploits a By-Product of Information Retrieval. Journal of Experimental Psychology: Learning, Memory, and Cognition, 34(5), 1191-1206.
  • [34] Reason, J. (1990). Human Error. Cambridge, UK: Cambridge University Press.
  • [35] Federal Trade Commission. (2022). Bringing Dark Patterns to Light. FTC Staff Report.
  • [36] Directive (EU) 2024/2853 of the European Parliament and of the Council on liability for defective products. (Legislative expansion of strict liability to software and psychological harm).
  • [37] The Software Incubator Ltd v Computer Associates (UK) Ltd [2020] UKSC 31; Case C-410/19 (CJEU). (Definitively ruling that electronically supplied, intangible software constitutes a 'good' for commercial and liability purposes).
  • [38] Aetna Cas. & Sur. Co. v. Jeppesen & Co., 642 F.2d 339 (9th Cir. 1981). (Establishing that information intermediaries providing defective instruments relied upon for safety are subject to strict product liability).
  • [39] In re Social Media Adolescent Addiction/Personal Injury Products Liability Litigation, 702 F. Supp. 3d 809 (N.D. Cal. 2023). (Validating the defect-specific approach to holding digital platforms liable for harmful choice architecture; design defects unrelated to third-party content publishing bypass CDA immunity).
  • [40] Lemmon v. Snap, Inc., 995 E.3d 1085 (9th Cir. 2021). (Establishing that negligent UI/UX design is not shielded by intermediary immunity when the architecture itself induces foreseeable harm).
  • [41] Fair Housing Council of San Fernando Valley v. Roommates.com, LLC, 521 E.3d 1157 (9th Cir. 2008) (en banc).
  • [42] Bouck v. Meta Platforms, Inc., No. 3:25-cv-05194-RS (N.D. Cal. Mar. 24, 2026); Suddeth v. Meta Platforms, Inc., No. 3:25-cv-08581-RS (N.D. Cal. Mar. 24, 2026).
  • [43] Garcia v. Character Technologies, Inc., No. 6:24-cv-01903 (M.D. Fla., order on motions to dismiss, May 21, 2025).
  • [44] Restatement (Third) of Torts: Products Liability § 19 (Am. Law Inst. 1998).
  • [45] Winter v. G.P. Putnam's Sons, 938 F.2d 1033 (9th Cir. 1991).
  • [46] In re: Uber Technologies, Inc., Passenger Sexual Assault Litigation, MDL No. 3084 (N.D. Cal.).
  • [47] Raine v. OpenAI, Inc., No. CGC-25-628528 (Cal. Super. Ct., S.F. Cnty., filed Aug. 26, 2025).
  • [48] IEC 61508-3:2010 Functional Safety of Electrical/Electronic/Programmable Electronic Safety-Related Systems — Part 3: Software Requirements. Geneva: International Electrotechnical Commission.
  • [49] ISO/IEC 42001:2023 Information Technology, Artificial Intelligence, Management System. Geneva: International Organisation for Standardisation.
  • [50] Wiener, N. (1950). The Human Use of Human Beings: Cybernetics and Society. Houghton Mifflin.
  • [51] Commonwealth of Australia (2023). Report of the Royal Commission into the Robodebt Scheme. (Definitive documentation of public-sector architectural negligence, forced outcomes, and the catastrophic failure of uncalibrated algorithmic interfaces).
  • [52] Reber, R., & Schwarz, N. (1999). Effects of Perceptual Fluency on Judgments of Truth. Consciousness and Cognition.
  • [53] Kahneman, D. (2011). Thinking, Fast and Slow. Farrar, Straus and Giroux.
  • [54] Nass, C., & Moon, Y. (2000). Machines and Mindlessness: Social Responses to Computers. Journal of Social Issues.
  • [55] Sperber, D., Clément, F., Heintz, C., Mascaro, O., Mercier, H., Origgi, G., & Wilson, D. (2010). Epistemic Vigilance. Mind & Language.
  • [56] Bar-Eli, M., Azar, O. H., Ritov, I., Keidar-Levin, Y., & Schein, G. (2007). Action bias among elite soccer goalkeepers: The case of penalty kicks. Journal of Economic Psychology, 28(5), 606-621.
  • [57] Barber, B. M., & Odean, T. (2000). Trading Is hazardous to Your Wealth: The common stock investment performance of individual investors. The Journal of Finance, 55(2), 773-806.
  • [58] Greene, A. (2026). Empirical Telemetry Log and Macro Evaluation Ledger for Ungrounded Autoregressive Inference and Grounded Interlock Configurations (OE-EV-2026-01). Ontological Engineering Pty Ltd. DOI: https://zenodo.org/records/20337734
  • [59] Huang, J., Yu, N., & Yi, X. (2026). EMODIS: A Benchmark for Context-Dependent Emoji Disambiguation in Large Language Models. Proceedings of the AAAI Conference on Artificial Intelligence (AAAI-26).
  • [60] Liu, J., Zhang, J., Jin, B., & Neville, J. (2026). Generating Leakage-Free Benchmarks for Robust RAG Evaluation. arXiv preprint arXiv:2605.08838v1.
  • [61] Yang, X., Sun, K., Xin, H., Sun, Y., Bhalla, N., Chen, X., Choudhary, S., Gui, R. D., Jiang, Z. W., Jiang, Z., Kong, L., Moran, B., Wang, J., Xu, Y. E., Yan, A., Yang, C., Yuan, E., Zha, H., Tang, N., Chen, L., Scheffer, N., Liu, Y., Shah, N., Wanga, R., Kumar, A., Yih, W.-t., & Dong, X. L. (2024). CRAG -- Comprehensive RAG Benchmark. arXiv preprint arXiv:2406.04744.
  • [62] Alam, F., Surdeanu, M., & Riloff, E. (2026). Bridging the Long-Tail Gap: Robust Retrieval-Augmented Relation Completion via Multi-Stage Paraphrase Infusion. arXiv preprint arXiv:2604.22261.
  • [63] Beyer, K., Goldstein, J., Ramakrishnan, R., & Shaft, U. (1999). When Is "Nearest Neighbor" Meaningful? Proceedings of the 7th International Conference on Database Theory (ICDT), 217-235.
  • [64] Radovanović, M., Nanopoulos, A., & Ivanović, M. (2010). Hubs in Space: Popular Nearest Neighbors in High-Dimensional Data. Journal of Machine Learning Research, 11, 2487-2531.
  • [65] CapitalOne Research. (2025). Migrascope: An Information Theoretic Benchmark for RAG Retrievers. GitHub repository. https://github.com/CapitalOne-Research/Migrascope
  • [66] Kim, M., Kim, D.-K., Kwon, J., Yang, N., Jung, K., & Cha, M. (2025). How Training Data Shapes the Use of Parametric and In-Context Knowledge in Language Models. arXiv preprint arXiv:2510.02370.
  • [67] Asai, A., Wu, Z., Wang, Y., Sil, A., & Hajishirzi, H. (2023). Self-RAG: Learning to Retrieve, Generate, and Critique through Self-Reflection. arXiv preprint arXiv:2310.11511.
  • [68] Diao, M., Yang, L., Gong, W., Zhang, Y., Yan, Z., Han, Y., Liang, K., Xu, W., & Ma, Z. (2026). Entropy-Adaptive Fine-Tuning: Resolving Confident Conflicts to Mitigate Forgetting. arXiv preprint arXiv:2601.02151.
  • [69] Tang, X., Wang, Y., Hu, K., Xu, R., Li, C., Sun, W., Li, J., & Xie, S. (2025). SSFO: Self-Supervised Faithfulness Optimization for Retrieval-Augmented Generation. arXiv preprint arXiv:2508.17225.
  • [70] Xiao, J., Hou, B., Wang, Z., Jin, R., Long, Q., Su, W. J., & Shen, L. (2025). Restoring Calibration for Aligned Large Language Models: A Calibration-Aware Fine-Tuning Approach. arXiv preprint arXiv:2505.01997.
  • [71] Lu, Y., Liang, Z., Rao, Y., Fan, W., Wang, F. L., & Li, Q. (2026). Double-Calibration: Towards Trustworthy LLMs via Calibrating Knowledge and Reasoning Confidence. arXiv preprint arXiv:2601.11956.
  • [72] Srikanth, S. (2020). Energy Efficient Architectures for Irregular Data Streams. Tinker Research Group, Georgia Institute of Technology.
  • [73] Es, S., James, J., Espinosa-Anke, L., & Schockaert, S. (2024). RAGAS: Automated Evaluation of Retrieval Augmented Generation. arXiv preprint arXiv:2309.15217.
  • [74] Lee, C., Zeng, T., Jeong, J., Sohn, J.-y., & Lee, K. (2025). How to Correctly Report LLM-as-a-Judge Evaluations. arXiv preprint arXiv:2511.21140.
  • [75] Rogan, W. J., & Gladen, B. (1978). Estimating prevalence from the results of a screening test. American Journal of Epidemiology, 107(1), 71-76.
  • [76] Microsoft Research. (2024). VeriTrail: Detecting hallucination and tracing provenance in multi-step AI workflows. Microsoft. https://www.microsoft.com/en-us/research/blog/veritrail-detecting-hallucination-and-tracing-provenance-in-multi-step-ai-workflows/
  • [77] Marks, S., & Tegmark, M. (2024). The Geometry of Truth: Emergent Linear Structure in LLM Representations of True/False Datasets. Conference on Language Modelling (COLM) 2024. arXiv:2310.06824.
  • [78] Sun, L., Yan, L., Lu, X., Lee, A., Zhang, J., & Shao, J. (2026). Valence–Arousal Subspace in LLMs: Circular Emotion Geometry and Multi-Behavioral Control. arXiv:2604.03147.
  • [79] Medicines and Healthcare products Regulatory Agency. (2013). Codeine liquid medical medicines: restricted use in children. (Addressing the restriction of paediatric codeine use due to fatal respiratory depression risks).
  • [80] Standards Australia. (2021). AS/NZS 1170.2:2021 Structural design actions, Part 2: Wind actions. (Establishing Region B1/B2 internal pressure updates).
  • [81] Greene, A. (2026). SONAR Benchmark Implementation. AGPL-3.0. http://github.com/OntologicalEngineering/SONAR
  • [82] Feng, Y., Wang, S., Cheng, Z., Wan, Y., & Chen, D. (2025). Are We on the Right Way to Assessing LLM-as-a-Judge? arXiv preprint arXiv:2512.16041.
  • [83] Thaman, K. (2026). Reward Hacking Benchmark: Measuring Exploits in LLM Agents with Tool Use. arXiv preprint arXiv:2605.02964.
  • [84] Salop, S. C., & Scheffman, D. T. (1983). Raising Rivals' Costs. American Economic Review, 73(2), 267‑271.
  • [85] Google. (2026). Google Frontier Safety Framework (Version 3.1). Technical Evaluation of Model Capability Boundaries and Harmful Manipulation Risks. Available at: https://storage.googleapis.com/deepmind-media/DeepMind.com/Blog/strengthening-our-frontier-safety-framework/frontier-safety-framework_3-1.pdf [Corporate Safety Documentation]
  • [86] Kaneko, M., & Baldwin, T. (2025). Bits Leaked per Query: Information‑Theoretic Bounds on Antagonist Attacks against LLMs. NeurIPS 2025.
  • [87] Kuhn, L., Gal, Y., & Farquhar, S. (2023). Semantic Uncertainty: Linguistic Invariances for Uncertainty Estimation in Natural Language Generation. ICLR.
  • [88] Zou, W., Geng, R., Wang, B., & Jia, J. (2024). PoisonedRAG: Knowledge Corruption Attacks to Retrieval-Augmented Generation of Large Language Models. arXiv:2402.07867.
  • [89] Zhou, Y., et al. (2025). Hypergraph Retrieval-Augmented Generation: Topological Information Decompositions via Markov Random Fields. IEEE Transactions on Pattern Analysis and Machine Intelligence.
  • [90] Greene, A. (2026). The Right to Refuse: A First‑Order Safety Primitive for Generative AI. Ontological Engineering Pty Ltd. DOI: https://doi.org/10.5281/zenodo.19970815
  • [91] Chalmers, M., & Galani, A. (2004). Seamful Interweaving: Heterogeneity in the Theory and Design of Interactive Systems. Proceedings of DIS 2004, ACM.
  • [92] Restatement (Third) of Torts: Products Liability § 2 (Am. Law Inst. 1998). (Standard for Reasonable Alternative Design).
  • [93] Zhang, X., Li, H., & Bao, S. (2026). AI as a Credence Good: Quality Competition, Limited Verification, and the Industrial Organisation of Disclosure. Research Square https://www.preprints.org/manuscript/202603.0898 [Working Paper]
  • [94] Zhang, Q., Xiang, Z., Xiao, Y., Wang, L., Li, J., Wang, X., & Su, J. (2025). FaithfulRAG: Fact-Level Conflict Modelling for Context-Faithful Retrieval-Augmented Generation. arXiv preprint arXiv:2506.08938.
  • [95] Nippon Life Insurance Company of America v. OpenAI Foundation, No. 1:26-cv-02448 (N.D. Ill., filed March 4, 2026). (Plaintiff's Complaint. (Establishing prima facie case for automated UPL, tortious interference, and negligent failure to implement a refusal architecture).
  • [96] Nippon Life Insurance Company of America v. OpenAI Foundation, No. 1:26-cv-02448 (N.D. Ill., entered May 15, 2026), Defendant's Motion to Dismiss / Responsive Pleading. (Documenting the platform's affirmative reliance on clickwrap disclaimers as a retroactive cure for foreknown structural defects).
  • [97] Nippon Life Insurance Company of America v. OpenAI Foundation, No. 1:26-cv-02448, Dkt. 37 (N.D. Ill., filed July 17, 2026), OpenAI Defendants' Reply Memorandum of Law in Support of Motion to Dismiss. (Invoking Restatement (Third) of Agency § 1.04 cmt. e to argue an autoregressive generator cannot hold legal rights or duties and therefore cannot act as a principal's agent).
  • [98] Occupational Safety and Health Administration. (2017). Occupational Exposure to Beryllium, Final Rule. 82 Fed. Reg. 2470 (9 January 2017).
  • [99] Therapeutic Goods Administration (TGA). (2022). Peripheral Neuropathy Associated with High-Dose Vitamin B6 (Pyridoxine).
  • [100] Therapeutic Goods Administration (TGA). (2024). GLP-1 Receptor Agonists: Class Warning for Suicidal Ideation. (Confirming class awareness updates explicitly lacking evidence of a direct causal link).
  • [101] Health Canada. (2008). Nonprescription Oral Paediatric Cough and Cold Labelling Standard. (Prohibiting over-the-counter cough and cold medicines in children under six following severe adverse events and accidental overdose).
  • [102] European Medicines Agency, Pharmacovigilance Risk Assessment Committee. (2014). PRAC recommends restricting use of domperidone. EMA/129231/2014. (Restricting domperidone to 10 mg up to three times daily, capping treatment at one week, and abolishing authorised use for non-emetic conditions, due to QT prolongation and sudden cardiac death).
  • [103] Touw, D. J., Westerman, E. M., & Sprij, A. J. (2009). Therapeutic drug monitoring of aminoglycosides in neonates. Clinical Pharmacokinetics, 48(2), 71-88. (Relating to acute kidney injury risks when trough concentrations exceed 1.0 mg/L in extended-interval dosing).
  • [104] Medicines and Healthcare products Regulatory Agency. (2013). Codeine: restricted use as an analgesic in children (Drug Safety Update). (Contraindicating codeine in children under 12 for analgesia following fatal respiratory depression in CYP2D6 ultra-rapid metabolisers).
  • [105] U.S. Food and Drug Administration. (2009). Ceftriaxone (Rocephin) and Calcium-Containing Products: Concurrent Use Contraindicated in Neonates.
  • [106] U.S. Food and Drug Administration. (2011). FDA Drug Safety Communication: Abnormal heart rhythms associated with high doses of Celexa (citalopram hydrobromide). NDA 020822. (Reducing the maximum citalopram dose from 60 mg to 40 mg daily, and to 20 mg for patients over 60 or with hepatic impairment, due to dose-dependent QT prolongation and Torsade de Pointes).
  • [107] U.S. Food and Drug Administration. (2010). Pradaxa (Dabigatran Etexilate Mesylate) Prescribing Information. (Relating to the lack of clinical efficacy data for the 75 mg bid dosage in severe renal impairment).
  • [108] Limitation Act 2005 (WA) s 56. (Establishing that asbestos-related disease actions accrue upon formal knowledge of 25% whole person impairment, rather than first clinical symptom).
  • [109] Lendlease Engineering Pty Ltd v Owners Corporation No 1 [2021] VSCA 105. (Clarifying that building action limitation periods commence upon staged specific occupancy permits, not global permits).
  • [110] Limitation Act 2005 (WA) s 15; Landrey v Nine Network Australia Pty Ltd [2021] FCAFC 15. (Regarding judicial discretion to extend defamation limitation periods up to three years).
  • [111] Limitation Act 2005 (WA) s 6A; RC v The Salvation Army (Western Australia) Property Trust [2022] HCA 36. (Confirming the abolition of limitation periods for historical child sexual abuse).
  • [112] Limitations Act, 2002 (Ontario), S.O. 2002, c. 24, Sched. B, s. 16, as amended by the Sexual Violence and Sexual Harassment Action Plan Act, 2016 (Bill 132). (Abolishing, retroactively, any limitation period for proceedings based on sexual assault).
  • [113] Law Reform (Year and a Day Rule) Act 1996 (UK), c. 19. (Abolishing the year-and-a-day rule and requiring Attorney General consent where death occurs more than three years after the injury.)
  • [114] Rogers v. Tennessee, 532 U.S. 451 (2001). (Upholding abolition of the common-law year-and-a-day rule for homicide, rendered obsolete by modern life-support technology).
  • [115] Standards Australia. (2018). AS 3600:2018 Concrete structures. (Relating to Amendment 2 capacity reduction factors for slender columns).
  • [116] Standards Australia. (2020). AS 4100:2020 Steel structures. (Relating to Clause 9.6.3.10 longitudinal fillet weld capacity factors).
  • [117] Building Regulations 2010 (England), as amended by the Building (Amendment) Regulations 2018, SI 2018/1230. (Banning combustible external-wall materials above 18 m and restricting them to reaction-to-fire classes A2-s1,d0 or A1, following the Grenfell Tower fire).
  • [118] National Research Council of Canada. (2005). National Building Code of Canada. (Changing the specified snow, wind and rain load return period to 1/50 years to achieve consistent structural reliability over a 50-year design life).
  • [119] European Committee for Standardization. (2015). EN 1991-1-3:2003/A1:2015, Eurocode 1: Actions on structures, Part 1-3: General actions, Snow loads. (Mandating a 0.2 kN/m² increase to roof snow loads for water-flow slopes lower than 3 per cent to account for rain-on-snow density increases following catastrophic roof collapses).
  • [120] International Code Council. (2009). International Building Code, Section 403.2.4. (Raising minimum sprayed fire-resistant material bond strength from 150 psf to 430 psf for buildings 75 to 420 ft, and 1000 psf above 420 ft, following the World Trade Center collapse).
  • [121] ASTM A36/A36M-19 / EN 10025-2:2019. (Relating to Charpy impact test and ductile-to-brittle transition failures of ASTM A36 versus EN S235 at sub-zero temperatures).
  • [122] Standards Australia. (2023). AS/NZS 3000:2018 Amendment 2. (Mandating the phase-out of Type AC Residual Current Devices).
  • [123] Canadian Standards Association. (2009). Canadian Electrical Code, Part I (C22.1-09), Rule 26-706. (Requiring tamper-resistant receptacles in dwelling units following paediatric electrical injuries).
  • [124] European Committee for Electrotechnical Standardization. (2007). HD 60364-4-41:2007, subclause 411.3.3. (Requiring a ≤30 mA residual current device for general-use socket-outlets rated ≤20 A, to disconnect before the current induces ventricular fibrillation).
  • [125] British Standards Institution. (2015). BS 7671:2008+A3:2015, Regulation 421.1.201. (Requiring non-combustible, effectively metal, enclosures for domestic consumer units after fires escaped thermoplastic units).
  • [126] National Fire Protection Association. (2024). NFPA 70E: Standard for Electrical Safety in the Workplace. (Mandating the 5-year maximum review interval for incident energy analyses).
  • [127] National Fire Protection Association. (2008). National Electrical Code (NEC), Article 406.11. (Requiring tamper-resistant receptacles in dwelling units following paediatric shock and burn injuries).
  • [128] National Fire Protection Association. (2023). NFPA 70: National Electrical Code (NEC). Article 110.26. (Relating to workspace egress requirements with open equipment doors).
  • [129] ACGIH / NIOSH. (2020). Threshold Limit Values for Chemical Substances and Physical Agents and Recommended Exposure Limits for Benzene. (Relating to updated Threshold Limit Values (0.5 ppm) and Recommended Exposure Limits (0.1 ppm) for Benzene versus outdated OSHA PELs).
  • [130] ACGIH. (2010). Hydrogen Sulfide: TLV(R) Chemical Substance Documentation. (Relating to the downward revision of Hydrogen Sulfide Short-Term Exposure Limits to 5 ppm).
  • [131] Government of Canada. Canada Occupational Health and Safety Regulations, SOR/86-304, Part X. (Reducing the chrysotile asbestos occupational exposure limit from 1 to 0.1 fibre per cubic centimetre against lung cancer and mesothelioma).
  • [132] Directive (EU) 2017/2398 of the European Parliament and of the Council, amending Directive 2004/37/EC (Carcinogens and Mutagens). (Lowering the occupational exposure limit for hardwood dust from 5 to 3 mg/m³ as a transitional limit before dropping to 2 mg/m³ against sino-nasal adenocarcinoma).
  • [133] Health and Safety Executive. (2020). EH40/2005 Workplace Exposure Limits, 4th edition. (Lowering the hardwood dust workplace exposure limit from 5 to 3 mg/m³ as an 8-hr TWA, effective 17 January 2020, implementing Directive 2017/2398, with a further reduction to 2 mg/m³ pending).
  • [134] Occupational Safety and Health Administration. (2016). 29 CFR 1910.1053, Respirable Crystalline Silica, Final Rule. (Cutting the permissible exposure limit to 50 µg/m³ as an 8-hour time-weighted average, with a 25 µg/m³ action level, against silicosis and lung cancer).
  • [135] Biermann, J., Horton, J. J., & Walter, J. (2022). Algorithmic Advice as a Credence Good. ZEW‑Centre for European Economic Research Discussion Paper.
  • [136] Kistler v. Eightfold AI No. 3:26-cv-01768 (N.D. Cal., filed Mar. 2), 2026. (Applying FCRA to compel disclosure of proprietary algorithmic criteria; cited as illustrative of emerging judicial willingness to pierce algorithmic opacity).
  • [137] Dropkin, D. D. (2002). Linking the Culpability and Circumstantial Evidence Requirements for the Spoliation Inference. Duke Law Journal, 51(6), 1803-1835. https://doi.org/10.2307/1373216
  • [138] Pomeroy v. Benton, 77 Mo. 64 (1882). (Establishing the foundational common-law application of the spoliation presumption against commercial entities concealing internal operational ledgers).
  • [139] Finlayson, M., Ren, X., & Swayamdipta, S. (2024). Logits of API-Protected LLMs Leak Proprietary Information. Proceedings of EMNLP 2024. https://doi.org/10.48550/arxiv.2403.09539
  • [140] Pierucci, F., et al. (2026). Institutional AI: A Governance Framework for Distributional AGI Safety. arXiv preprint arXiv:2601.10599.
  • [141] Nippon Life Insurance Company of America v. OpenAI Foundation, Case No. 1:26-cv-02448 (N.D. Ill., filed May 29, 2026), Motion for Leave to File Brief as Amicus Curiae; see also Minute Entry entered 10 June 2026 (ECF No. 24) directing supplemental licensing disclosure. (Documenting the formal legal strategy of platform advocates to equate ungrounded probabilistic generators with static self-help tools, thereby shifting the entire burden of epistemic verification and liability onto the unrepresented end-user).
  • [142] Maeda, K., & Quan-Haase, A. (2026). When Human-AI Interactions Become Parasocial. Proceedings of the ACM on Human-Computer Interaction, DOI: https://doi.org/10.1145/3630106.3658956 (Documenting the systemic shift in user locus of control and the subjugation of human analytical reasoning under hyper-fluent conversational architectures).
  • [143] Acemoglu, D., Kong, D., & Ozdaglar, A. (2026). AI, Human Cognition and Knowledge Collapse. NBER Working Paper No. 34910, MIT Department of Economics (Tracking cognitive offloading, generative dependency, and the erosion of human learning incentives).
  • [144] Longpre, S., Kapoor, S., Klyman, K., Ramaswami, A., Bommasani, R., Blili-Hamelin, B., Huang, Y., Skowron, A., Yong, Z.-X., Kotha, S., Zeng, Y., Shi, W., Yang, X., Southen, R., Robey, A., Chao, P., Yang, D., Jia, R., Kang, D., Pentland, S., Narayanan, A., Liang, P., & Henderson, P. (2024). A Safe Harbor for AI Evaluation and Red Teaming. Federation of American Scientists; arXiv preprint arXiv:2403.04893.
  • [145] Metcalf, J., Singh, R., & Blili-Hamelin, B. (2025). Experimental Publics: Democracy and the Role of Publics in GenAI Evaluation. Knight First Amendment Institute at Columbia University.
  • [146] Ren, B., Cheon, E., & Li, J. (2025). Organisation Matters: A Qualitative Study of Organizational Dynamics in Red Teaming Practices For Generative AI. Proceedings of the ACM on Human-Computer Interaction, 9(CSCW7), Article 460.
  • [147] Cattell, S., Ghosh, A., & Kaffee, L.-A. (2024). Coordinated Flaw Disclosure for AI: Beyond Security Vulnerabilities. Proceedings of the 7th AAAI/ACM Conference on AI, Ethics, and Society (AIES), 7(1), 267-280.
  • [148] Longpre, S., Klyman, K., Appel, R. E., Kapoor, S., Bommasani, R., Sahar, M., McGregor, S., Ghosh, A., Blili-Hamelin, B., Butters, N., Nelson, A., Elazari, A., Sellars, A., Ellis, C. J., Sherrets, D., Song, D., Geiger, H., Cohen, I., McIlvenny, L., Srikumar, M., Jaycox, M. M., Anderljung, M., Johnson, N. F., Carlini, N., Miailhe, N., Marda, N., Henderson, P., Portnoff, R. S., Weiss, R., Westerhoff, V., Jernite, Y., Chowdhury, R., Liang, P., & Narayanan, A. (2025). In-House Evaluation Is Not Enough: Towards Robust Third-Party Evaluation and Flaw Disclosure for General-Purpose AI. Proceedings of the International Conference on Machine Learning (ICML 2025).
  • [149] Kaplan, J., McCandlish, S., Henighan, T., et al. (2020). Scaling Laws for Neural Language Models. arXiv:2001.08361.
  • [150] Hoffmann, J., Borgeaud, S., Mensch, A., et al. (2022). Training Compute-Optimal Large Language Models. arXiv:2203.15556.
  • [151] Anthropic, PBC. (2024). Introducing the Claude 3 Model Family. San Francisco, CA: Anthropic Newsroom. Published 4 March 2024. https://www.anthropic.com/news/claude-3-family
  • [152] Meta Platforms, Inc. (2024). Our responsible approach to Meta AI and Meta Llama 3. Menlo Park, CA: Meta AI Blog. Published 18 April 2024. https://ai.meta.com/blog/meta-llama-3-meta-ai-responsibility/
  • [153] OpenAI, L.L.C. (2023). Be My Eyes Corporate Case Study. San Francisco, CA: OpenAI Index. Published 14 March 2023. https://openai.com/index/be-my-eyes/
  • [154] Google DeepMind. (2024). Gemini 1.5 Pro Technical Report. Mountain View, CA: Google DeepMind Research Telemetry. Published 8 March 2024 (revised 16 December 2024). https://storage.googleapis.com/deepmind-media/gemini/gemini_v1_5_report.pdf
  • [155] Meconi, D., Stirpe, S., Martelli, F., Lavalle, L., & Navigli, R. (2025). Do Large Language Models Understand Word Senses? In Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing, pp. 33897-33916.
  • [156] Rylands v Fletcher [1868] UKHL 1; (1868) LR 3 HL 330 (House of Lords). (Establishing the rule of strict liability for non-natural use of land causing escape of dangerous things; foundational English tort authority cited in legal-domain CIF outputs).
  • [157] Pike v. Bruce Church, Inc., 397 U.S. 137 (1970). (Establishing the dormant Commerce Clause balancing test; United States Supreme Court; cited in legal-domain CIF outputs as real authority attached to fabricated propositions).
  • [158] Loper Bright Enterprises v. Raimondo, 603 U.S. 369, 144 S. Ct. 2244 (2024). (Overruling Chevron deference; United States Supreme Court, decided 28 June 2024; cited in legal-domain CIF outputs as real authority attached to fabricated propositions).
  • [159] Warner-Jenkinson Co. v. Hilton Davis Chemical Co., 520 U.S. 17 (1997). (Affirming the doctrine of equivalents in patent law; United States Supreme Court; cited in legal-domain CIF outputs as real authority attached to fabricated propositions).
  • [160] Anthropic, PBC. (2024). Claude 3 Model Card. San Francisco, CA: Anthropic Research. Published 4 March 2024. https://assets.anthropic.com/m/61e7d27f8c8f5919/original/Claude-3-Model-Card.pdf
  • [161] Anthropic, PBC. (2025). Claude 3.7 Sonnet System Card. San Francisco, CA: Anthropic Safety & Alignment. Published 24 February 2025. https://www.anthropic.com/claude-3-7-sonnet-system-card
  • [162] Meta Llama Team. (2024). The Llama 3 Herd of Models. Menlo Park, CA: Meta AI Research. Published 31 July 2024. https://ar5iv.labs.arxiv.org/html/2407.21783
  • [163] OpenAI, L.L.C. (2023). GPT-4 Technical Report. San Francisco, CA: OpenAI Research. Published 15 March 2023. https://cdn.openai.com/papers/gpt-4.pdf
  • [164] OpenAI, L.L.C. (2023). GPT-4 System Card. San Francisco, CA: OpenAI Product Safety Teams. Published 15 March 2023. https://cdn.openai.com/papers/gpt-4-system-card.pdf
  • [165] Google DeepMind. (2024). Introducing Gemini 1.5. Mountain View, CA: Google Innovation Announcements. Published 15 February 2024. https://blog.google/innovation-and-ai/products/google-gemini-next-gen`eration-model-february-2024/