Ronke Jegede.

Munir v Secretary of State for the Home Department [2026] UKUT 81 (IAC) — the Upper Tribunal ruled at paragraph 60 that uploading confidential client documents to open-source AI tools such as ChatGPT places material in the public domain, breaching confidentiality and permanently waiving legal professional privilege. Intention is irrelevant. The waiver is irreversible.

The regulatory obligations that follow: The Tribunal stated that such conduct may warrant referral to the SRA and must be referred to the ICO. The word "must" in relation to ICO notification is not discretionary. Your firm needs a written incident response process that reflects both obligations — with a 72-hour ICO notification clock running from discovery.

Cork and another v Mark Smith [2026] EWHC 1199 (Ch) — ICC Judge Mullen, 22 May 2026. AI-hallucinated Insolvency Rules went to court at Pinsent Masons unchecked. The firm self-referred to the SRA. The judgment reinforced that the supervising solicitor retains full responsibility — and may be more culpable than the junior who introduced the AI error. A lack of supervision protocol is not a defence. It is an aggravating factor.

The governing distinction from Munir: The Tribunal drew an explicit line between open-source public tools (privilege waiver risk) and secure closed enterprise systems with appropriate safeguards (lower risk). Every AI tool at your firm must be classified against this distinction immediately. Anything in the public category must be restricted from use on client matters.

The ten-point governance response: Audit your AI tool inventory. Classify every tool as open or closed. Obtain and review vendor DPAs. Name a governance owner for each AI deployment. Write a human review protocol for AI outputs. Update client disclosure language. Establish an AI incident reporting process. Train all fee earners on the Munir distinction. Brief the partnership on the Cork supervision point. Document everything with an audit trail.

The EU AI Act operates on a territorial basis that catches organisations regardless of where they are incorporated. Article 2 of the Act applies it to any provider or deployer whose AI system output is used in the EU — meaning a London law firm advising a Frankfurt corporate client through an AI-assisted research tool is in scope.

What hits in August 2026: Article 50 transparency obligations require explicit disclosure when AI is used in client-facing interactions. Article 4 requires documented AI literacy training for all staff using AI tools. Neither obligation was extended by the May 2026 Omnibus — the high-risk Annex III deadline moved to December 2027, but Article 50 and Article 4 did not.

Which firms are most exposed: Any firm with EU clients, EU offices, or EU-facing AI deployments. Magic Circle and large regional firms with Brussels or Frankfurt desks have the most immediate obligations. Mid-size firms handling cross-border M&A, arbitration, or regulatory matters for EU corporates are the most overlooked category.

The practical steps: Conduct an AI tool inventory and classify each tool's EU-facing use. Implement client disclosure language in engagement letters. Deliver and document Article 4 AI literacy training. These are not complex governance tasks — but they need to be done before August 2.

What shadow AI looks like in practice: A solicitor drafts a client letter using personal ChatGPT. A trainee summarises a brief using Gemini on their personal laptop. A partner runs opposing counsel's submissions through Claude to identify weaknesses. None of these uses are firm-sanctioned, none have Data Processing Agreements, and none are disclosed to the client. All three are live UK GDPR violations and potential SRA Code breaches.

The SRA exposure: SRA Code of Conduct for Solicitors, Paragraph 3.5 holds supervising solicitors personally accountable for all work carried out under their supervision — including AI-assisted work. A supervising partner who cannot demonstrate oversight of how their team uses AI on client matters faces personal regulatory action. The firm faces institutional sanction under Codes for Firms Rules 2.1(a), 4.2, 4.3, and 4.4.

The UK GDPR exposure: Client data entered into personal AI accounts without a Data Processing Agreement (Article 28) and without a valid lawful basis (Article 6) is a live ICO enforcement risk. The ICO has made AI data protection one of its 2025–2026 enforcement priorities. Enforcement action does not require a data breach — a compliance audit finding shadow AI without DPAs is sufficient.

What adequate governance looks like: A documented AI tool inventory. A firm-wide AI Acceptable Use Policy. Data Processing Agreements with all AI vendors. Client disclosure language in engagement letters. Staff training records. None of this is technically complex — but all of it needs to exist before the next SRA audit.

The distinction that matters: A supervised AI tool produces output that a human reviews before anything happens. An AI agent takes actions — sending emails, accessing databases, generating documents, scheduling tasks — without requiring a human to initiate each step. This is not a marginal technical difference. It is a fundamental shift in where the accountability sits.

Why existing frameworks fail: Most law firm AI policies regulate AI outputs — they require human review of AI-generated documents. Agentic AI acts between those review points. A research agent that browses legal databases and assembles a case analysis is doing work that no existing policy framework regulates, because the policy was written for a tool that produces a draft, not for a system that autonomously assembles evidence.

The six governance elements agentic AI requires: First, explicit autonomy boundary definition — what the agent can do without human approval. Second, mandatory human-in-the-loop checkpoints for high-stakes actions. Third, comprehensive audit logging of all autonomous actions. Fourth, clear liability allocation between the firm, the supervising lawyer, and the AI vendor. Fifth, client disclosure that an agent — not just an AI tool — is involved in their matter. Sixth, incident response protocols specifically for agentic failures, hallucinations, and boundary violations.

The practical starting point: Before deploying any agentic AI system on client matters, define the human oversight architecture first. Every autonomous action the agent can take should be explicitly listed, risk-classified, and assigned a human accountability owner. The governance framework shapes the deployment — not the other way around.

What the SRA guidance actually says: The SRA's Technology Guidance on AI (2024) sets out expectations across four areas: competence (SRA Code 1.3 — solicitors must understand AI well enough to supervise its outputs), oversight (Para 3.5 — personal accountability for AI-assisted work), client disclosure (Principle 4 and Code 8.6 — clients must be informed when AI materially affects their matter), and data protection (Code 4.2 — firms must have adequate systems for handling client information through AI tools).

What documented compliance looks like: A written AI governance policy reviewed within the last 12 months. A staff AI literacy training programme with attendance records. Engagement letter clauses disclosing AI use to clients. Data Processing Agreements with all AI vendors handling client data. A named individual with responsibility for AI governance. A log of AI tools in active use, reviewed quarterly.

What the SRA is looking for in practice: The SRA has signalled that AI governance will be included in thematic reviews of law firm operations. They are not expecting perfection — they are expecting evidence of deliberate governance. A firm that can produce a current AI policy, training records, and vendor DPAs will pass scrutiny. A firm that cannot produce any of these will not.

The ten-point compliance checklist: (1) Written AI Acceptable Use Policy — current. (2) AI system inventory — all tools documented. (3) Named AI governance owner. (4) Staff training records — all fee earners. (5) DPAs with AI vendors — signed and current. (6) Client disclosure language — in engagement letters. (7) Shadow AI controls — technical or procedural. (8) Human oversight protocols — documented per tool type. (9) AI incident reporting process — written and tested. (10) Board-level AI risk reporting — at least quarterly.

Why this library exists: Every firm deploying AI believes the headline failures happen to other firms. They do not. The Mata v. Avianca fabricated citations case, the Samsung data breach, and the UK employment tribunal AI research failure all began with the same assumption — that AI tools were sufficiently reliable to use without governance architecture. They were not.

The pattern across all 20 cases: No documented AI governance policy. No human oversight protocol for AI outputs. No client disclosure. No staff training records. No vendor due diligence. The governance failures are identical across jurisdictions, practice areas, and firm sizes. Only the technology and the consequences differ.

The August 2026 implication: From August 2, 2026, EU AI Act Article 50 transparency obligations require firms to disclose AI use to clients. A firm that cannot produce a governance framework demonstrating deliberate AI oversight is not just reputationally exposed — it is regulatory enforcement-exposed. The incidents in this library are the before. Your firm's governance programme is the after.