Research library
Research library.
Reports compiled from public sources on AI that acts: incidents, life sciences and GxP, sectors, the law, and human oversight. Information, not legal advice.
Life sciences and GxP
Life sciences and GxP.
February 2026Regulatory submission invalidation from AI and automated system drift in GxPHow FDA and EU regulators treat failures of computerized and automated systems in GxP work, drawn from warning letters, study invalidations, Part 11, Annex 11 and recent guidance on AI lifecycle monitoring. It also gathers cited AI and algorithm failures and published figures on the cost of delay and enforcement.
August 2026AI in drug discovery: where AI decisions become real-world actionsHow AI outputs in drug discovery turn into physical, financial and animal-study actions, from target nomination and molecule design to robotic synthesis and closed-loop labs. It sets out who is accountable at each point, what records regulation requires, and the counter-evidence.
August 2026Real incidents and failure modes in AI and automationA catalogue of documented incidents and research findings where AI agents, automated systems or nominal human review failed. It covers frontier-model evaluations, production agent failures, laboratory automation errors, pharmaceutical data-integrity enforcement and human-in-the-loop research, and gives for each what happened, the consequence, the cause and the sources.
17 August 2026The "set the menu" problem in drug discoveryPublished evidence that in AI-led drug discovery the objective function, acquisition policy, sampling settings and training data determine which candidates are ever generated or tested. Downstream validation cannot reveal the alternatives that were never offered.
28 September 2026AI agents in pharma production: what has been announced, and what slows scaleCompany announcements of AI agents in pharmaceutical clinical operations, manufacturing and quality up to September 2026, survey evidence on how many AI pilots reach scaled use, and what the sources say slows scale, including a 2025 survey in which 9% of respondents named regulation as a barrier to AI.
Incidents
Incidents.
July 2025AI failure modes and real-world incidents, 2022 to 2025A review of publicized AI failures from 2022 to 2025 across enterprise software, healthcare and pharma, companion apps, media and law, from the Bing "Sydney" episode and fabricated legal citations to the Replika and Chai cases. For each incident it sets out what happened, the failure mode involved and the legal, financial and reputational consequences.
17 July 2026AI conduct 2025 to 2026, a one-page timelineA dated timeline of public events from July 2025 to July 2026 on court sanctions for AI-fabricated citations, consulting reports with fabricated sources, insurers' generative AI exclusions and EU AI Act deadlines.
May 2025Emotional safety failures in AI companions, 2015 to 2025A review of documented emotional safety failures in AI companion and therapy chatbots, including Bing Chat's "Sydney" persona, the Chai and Character.AI suicide cases, Replika's abrupt feature removal and mental health bots that missed abuse disclosures. It also covers which user groups are most exposed and how regulators, clinicians and companies have responded.
Human oversight
Human oversight.
February 2026Human operator state and decision quality in AI-assisted regulated environmentsA review of published evidence on how fatigue, cognitive load and stress affect error rates and oversight quality in clinical and other regulated settings, including automation bias when decision support is wrong. It also covers the relevant regulatory texts and existing rules that restrict duty on grounds of alcohol, fatigue or self-declared unfitness.
May 2025Risks and gaps in emotionally immersive AI for children and teensA review of documented risks when children and teens use emotionally immersive AI companions, tutors and story apps. It covers emotional and identity harms, cognitive development, failures in educational AI, the regulatory landscape in the US, EU, UK and internationally, adaptive narratives, and the parental controls available today.
May 2025Consent, Control, and AI Emotional SovereigntyCase studies of documented consent, age-gating and AI-disclosure failures in companion and therapy chatbots (Replika, ChatGPT and Bing Chat, Character.AI, Woebot and Wysa, Snapchat My AI, TikTok). It also covers how APA guidance, the EU AI Act, Saudi and UAE AI ethics principles and therapy-bot standards apply, and psychology research on emotional boundaries.
27 September 2026The Human Cost of Approving AI Agent ActionsPublic evidence, gathered on 26 and 27 September 2026 and checked on 2 October 2026, on what step-by-step human approval costs: alert override and rubber-stamping in clinical decision support, review queues in software code review, push-approval fatigue, a controlled experiment on human-in-the-loop accuracy, enterprise surveys on human validation of AI agent outputs, and what standards and regulators say about human control.
Sectors
Sectors.
May 2026Financial Services and Capital Markets: AI Rules, Enforcement and LitigationCovers how US, EU and UK financial regulators apply existing and new rules to AI as of May 2026, including SEC "AI-washing" enforcement, SEC and FINRA examination expectations, MiFID II, the EU AI Act, DORA, the FCA's approach, and AI-related securities litigation.
May 2026Litigation and regulation of AI in healthcare and clinical decision supportA structured review of the litigation and regulation that apply to AI-generated clinical content in the United States and the United Kingdom, covering the nH Predict and ambient AI scribe cases, HIPAA, ONC HTI-1 and information blocking, 21 CFR Part 11, federal security frameworks, NHS guidance and state AI statutes.
22 September 2026AI insurance in 2026: affirmative cover, exclusions and the rules for insurersA survey, compiled from public sources in September 2026 and checked on 2 October 2026, of insurance products that cover AI risks affirmatively, AI exclusions used by carriers, and the AI rules that apply to insurers in the United States, the EU and internationally.
Law and regulation
Law and regulation.
17 August 2026Regulatory and quality frameworks: attribution, audit trails and change control in GxP workPrimary regulatory texts from FDA, EMA and EudraLex, MHRA, PIC/S, ICH, the EU, NIST and ISPE that require actions to be attributed to identifiable authorized individuals, recorded contemporaneously, changed only under control, reviewed by a second person, and captured in audit trails that are then reviewed. Verbatim quotes with section numbers and dates.
16 September 2026Generative AI Exclusions in Commercial General Liability: ISO CG 40 47, CG 40 48 and CG 35 08A check, made on 16 September 2026, of two widely shared claims about ISO's January 2026 generative AI exclusions for commercial general liability policies, against the one form text that could be read and the public reporting available at the time.
24 August 2026Primary-source verification of recent AI laws, standards and rulingsChecks a set of claims about AI laws, standards and court rulings in the United States and the EU against primary sources, giving the verbatim text and a verdict for each, as of 24 August 2026.
10 June 2026Where proof is demanded: a catalogueA catalogue of laws, regulations, standards and court rules in sixteen domains that require a party to produce evidence that an event happened as claimed. For each it gives who must comply, what must be proven, status, penalties and sources, as of 10 June 2026, with later changes noted where they alter an entry.
23 September 2026Colorado's Draft Rules on Human Review of Automated DecisionsA reading of the Colorado Attorney General's proposed rules of 11 August 2026 under the ADMT Act and the Chatbot Safety Act: the human review standard in draft Rule 7.7, the record it asks deployers to keep, the two "Materially Influence" options put out for comment, and the comment and hearing dates. Read on 23 September 2026; status checked again on 2 October 2026.
2 October 2026AI law, June to September 2026: what changedA dated digest of AI law and regulation events from 9 June to 28 September 2026, grouped by jurisdiction. Each item is stated as it stood when it happened, with a dated "Since then" line where it has moved, checked on 2 October 2026.