The Learning Center, by chapter

Chapter 1

Legal practice and the courts

Courts were first to write the duty down: verify AI work, under human control, and be ready to show it. This chapter holds the duty, the sanctions that price it, and the page for the lawyer who has to answer a judge on Monday.

The AI Review Receipt for attorneysWhat the human reviewed, signed, sealed and checkable by the court. Verification is becoming a dutyIreland's HC 142, India's Supreme Court and the Illinois Appellate Court, and the layer they leave missing. Ireland, Practice Direction HC 142In operation 1 September 2026. The verification duty, paragraph by paragraph. Caribbean Court of Justice, Practice Direction No. 1 of 2025A final appellate court reserving the power to ask how the verification was conducted. February 2025. Jamaica, Practice Direction No. 1 of 2025The mandatory declaration that names the AI tool and its version. September 2025. China, Supreme People's Court Opinions on AI disputesNational judicial guidance: verify before filing, explain the AI assistance on filing. September 2026, working translation. State ex rel. Schnell v. RichardsonAI use was not the violation. Filing unverified work was. Missouri Court of Appeals, September 2026. State v. ColemanFabricated transcript quotations, eight sanctions, and a signed verification certification on every filing for two years. Ohio, March 2026. The sanctions recordVerified sanction matters, penalties and patterns, including the cases that cut the other way. Standing ordersThe judges' standing orders on AI, in the orders' own language, with last-read dates. Court rulesThe court rules that reach AI use in filings. AI and privilegeThe split every litigator should know, and HC 142 paragraphs 10 and 32. When the evidence is AIDated records of AI-made evidential material.

Chapter 2

Life sciences and GxP

Batch release, data integrity and the audit trail were built for people. AI now writes into the same record. This chapter holds the pages for the quality unit, the Qualified Person and the validation lead.

AI-assisted science under human authorityWhere the check sits in a regulated lab, and what the record shows an inspector. The enterprise page. Annex 11 and AIWhat the audit trail has to show when AI touches a computerised system. EU GMP Annex 16Qualified Person batch release, and what the QP must be able to stand behind. EU GMP Annex 22, draftThe first GMP annex written for AI, read at the draft. FDA 21 CFR 211.22Quality control unit authority. The authority the regulation names. FDA 21 CFR Part 11Electronic records and electronic signatures. MHRA GxP data integrityALCOA+ and audit trails, at the guidance. PIC/S PI 041-1Data management and integrity for inspectors. ISPE GAMP Guide: Artificial IntelligenceThe validation profession's own AI guide. FDA AI credibility assessment, draftCredibility of AI for regulatory decisions, read at the draft. EMA reflection paper on AIAI across the medicinal product lifecycle. Applied Therapeutics: deleted audit trailsWhat an FDA inspection found when the trail was gone. The record. FDA warning letter: PuroleaThe record. Immensa: the PCR reporting thresholdA threshold change nobody signed. The record.

Chapter 3

Agents and authority

For security, identity, platform and model-risk teams. Agents act. The question that follows every action is who authorized it, and whether an outsider can check. This chapter holds the pages on agents that act under human authority, the records of agents and automated systems that did not, and the rules that now ask for the log.

Agentic AI under human authorityWhat an agent is allowed to do, who said so, and the record left behind. The enterprise page. What does the system prove, and who must be trusted?The four questions to put to any agent platform before it acts on your behalf. Replit and SaaStr: the deleted production databaseAn agent acted through an instruction freeze. The record. AWS Kiro: delete and recreateAn agent's remedy became the outage. The record. UK AISI: unsanctioned agent actionsAgents under evaluation stepped outside the sanctioned set. The record. Agents on three dormant German wikisAutonomous edits to sites nobody was watching. The record. EU AI Act, Article 12Automatic event logging for high-risk systems, in the Act's own words. EU AI Act, Articles 9, 13, 14 and 15Risk management, transparency, human oversight and robustness. Model risk and staff AI useWhat evidence supervisors ask for when AI is in the workflow. SR 26-2 and OCC Bulletin 2026-13Model risk management, the current supervisory statement, with the RFI status as at the last read. GDPR Article 22Evidencing meaningful human intervention in an automated decision. Knight CapitalForty-five minutes of unauthorized trading. The record. Moonwell: the cbETH oracleAn oracle input nobody checked before it moved funds. The record. RobodebtAutomated debt notices at national scale, and the royal commission that followed. The record.

Chapter 4

Insurance

Insurers meet AI twice: in their own claims and underwriting decisions, and in the conduct of the insureds they price. This chapter holds both.

Telematics for AI conductA record of what the AI did and what the human authorized, as underwriting evidence. NAIC AI Model BulletinWhat state regulators expect of an insurer's AI systems program. Colorado SB 26-189Automated decision transparency for consumers. Cigna PxDxClaims denied in seconds, and the litigation that followed. The record. UnitedHealth nH PredictAn algorithm in post-acute care decisions. The record. Moffatt v. Air CanadaA chatbot's answer bound the company. The record.

Chapter 5

The Consequence Library

Documented cases in which AI or automated systems took part in a consequential action. Each record asks the same four questions: what happened, where control failed, who held the authority, and what could be proven afterward.

The Consequence LibraryEvery record, filterable by sector, system type and failure stage. How records are made and gradedThe source standard, the three kinds of statement, the gap codes and the confidence grades.

Chapter 6

The laws

The full library, every sector, every country, each entry read at the issuing body's own document. Start from the law, or start from your problem and get routed to it.

The 146 lawsThe AI laws and rules for professional work, indexed, each with its source. AI rules by countryThe rules for AI in professional work, country by country. Pick the problemStart from your problem and get routed to the law and the proof. The Drift DeskThe week in AI, read through the record. Local Law 144The bias audit and the records you need to keep. When the law changesDoes a record sealed today still hold tomorrow? DORA, Regulation 2022/2554Digital operational resilience for the EU financial sector.

Reading the marks

Gold: the law and the landscape. Statutes, rules, guidance and where the ground is moving. Read at the issuing body's own document.
Navy: the record and the court. Orders, decided matters, consequence records and the instruments that answer them. What a judge, an auditor or an opponent actually reads.

Chapter 7

How Regulayer works

AI agents cannot be trusted to police themselves. Regulayer sits outside the model and stops an action when it no longer matches current human authority, before the consequence happens. Then it leaves a record an outsider can verify. These pages show it, and let you try to break it.

Why the authority sits outside the AIThe position, and why it cannot be talked around from inside the model. What is a Regulayer Receipt?The signed record of AI use and human review. The Receipt, in detailProve what AI did and what the human reviewed. HumanMarkHow the work was actually made, recorded at creation. Live RegulayerDon't take our word for it. Try to break it. Verify it yourselfCheck a Regulayer record without trusting Regulayer. Pick what it doesWatch it go from trust me to proof. Why nowThe gas pedal has had years. The brake is the new part.

The law moves. This library moves with it.

SignalDrop is the update: when a rule changes, a court decides, or a page in this library is re-read at source, it goes out as a concise summary with the primary source attached. No newsletter filler, and nothing you have to be a lawyer to read.

One address, nothing else. We do not sell it, we do not profile it, and every update carries a one-click unsubscribe.

Pick your chapter.