Skill Diligence

AI for In-House Counsel

Surprisingly Useful News from the Legal AI Frontier

Issue 27 · 29 June 2026

The US Just Proved Your AI Vendor's Access Can Vanish Overnight

Plus: Brussels delays its toughest AI Act obligations by 16 months, and a regulated AI law firm wins its first contested trial in a UK county court.

Here's what mattered most to me this week — an export-control directive can cut off your AI vendor access faster than any outage ever could, and most legal teams have no contingency plan for it. That's the one to act on now. Also in this issue: the European Parliament votes through a 16-month delay to the AI Act's toughest obligations, and an SRA-regulated AI law firm wins its first courtroom trial against barrister-led opposition.

1

This Week's Reading

Top Story
U.S. Department of Commerce / Bloomberg Law · 12 June 2026
Washington Orders Anthropic to Cut Off Foreign Access to Its Most Powerful Models — This Is Why You Need a System That Is Independent of Your AI Provider
If you're reading this from outside the US, you were the access this directive was written to remove. On 12 June, the US Commerce Department ordered Anthropic to suspend all foreign-national access to its two most advanced models, Fable 5 and Mythos 5 — a category that includes every lawyer, client and business outside America, regardless of how the tool was being used. Anthropic disabled both models worldwide while it worked to comply, citing a suspected jailbreak as the national-security trigger; a customer has since sued the US government over the lost access.
Key Insight:

Build your own operating system for AI-assisted legal work, not a workflow bolted onto someone else's platform. As counsel outside the US, you had no input into this decision and no recourse when it landed. Treat any single AI vendor as a component you could lose overnight, not infrastructure you depend on, and design your own processes so they survive losing any one provider. I wrote about this after Fable's suspension — read it here.

European Parliament · 16 June 2026
Brussels Just Gave High-Risk AI Systems 16 More Months Before the AI Act Bites
The toughest AI Act obligations just moved from August 2026 to December 2027 — a reprieve, not a repeal. On 16 June, the European Parliament voted 423 to 57 to approve amendments delaying application of the AI Act's high-risk system obligations, with embedded safety-component systems pushed back further still, to August 2028. Parliament also confirmed a new ban on "nudifier" apps that generate non-consensual intimate or child sexual abuse imagery, taking effect this December.
Key Insight:

Use the extra time to get ahead of the requirements, not to forget about them. If your organisation, or a client, operates a high-risk system caught by Annex III — employment, credit scoring, critical infrastructure — you now have until December 2027 rather than August 2026 to complete conformity assessments and risk-management documentation. Build that deadline into your 2027 planning now, while it still feels far away.

Garfield AI / Law Gazette · 22 June 2026
A Fully AI-Run Law Firm Just Won Its First Courtroom Trial in the UK
The claimant paid roughly £400 in AI legal fees to recover a £7,000 debt — the defendant's solicitor and barrister cost far more, and still lost. On 22 June, Garfield AI, the world's first SRA-regulated AI law firm, confirmed that a freelancer it represented won a three-hour trial at Wandsworth County Court on 14 May, recovering unpaid fees from a hospitality business and defeating a counterclaim. Both sides were represented by barristers; only one side's advocate was an AI system.
Key Insight:

Don't wait for a contested dispute to test whether AI-run advocacy can survive cross-examination — a UK court just did. Garfield is regulated for debt recovery only, but the precedent reaches further: a court has now accepted AI-led litigation conduct under SRA supervision and found in its favour. If your organisation handles low-value disputes or debt recovery in-house, ask whether a regulated AI provider could now do that work at a fraction of the cost.

2

This Week's Workshop

How to Build a Contract Triage Process Your Clients Can Run Themselves

Most in-house teams don't get buried by complex matters — they get buried by the same low-value, low-risk contract questions, asked over and over by people who don't need a lawyer's judgement, just a clear process. In this workshop, we build that process: you set the risk parameters once, AI turns them into a structured triage workflow, and that workflow becomes a page and form your client or business team can run themselves for anything below your threshold — anything above it still comes to you. Upload it to your intranet and watch the trivial queries start routing themselves.

Agent Builder Course — Five Days, One Hour a Day

Learn to build, deploy and continuously improve AI agents for your legal team's hardest recurring work — contract triage, due diligence, document assembly, compliance monitoring. No coding required. Works with any LLM, including Microsoft Copilot. Spaces are limited.

More details and booking here →

Book a Call

Copy-Paste Ready: The Contract Triage Process Builder Prompt

Use this prompt when you want to design a self-serve triage process for a recurring category of low-risk work — contract review, routine queries, standard approvals — so your client or business team can handle the low-value cases themselves, while anything above your risk threshold still reaches you. Paste it into Claude, ChatGPT, or any LLM, replace the bracketed details, and work through each section:

# CONTRACT TRIAGE PROCESS BUILDER PROMPT ## ROLE You are an experienced in-house legal operations consultant and process designer. Your task is to design a complete, self-serve triage process that lets a client or business team handle low-value, low-risk matters themselves, while routing anything above an agreed risk threshold to a lawyer. Be specific and operational. Every output should be something that could be published to an intranet or handed to a business team this week — not a description of what a good process would look like in general. --- ## INPUT **Contract or matter type:** [INSERT TYPE — e.g. NDAs, vendor agreements, low-value purchase orders] **Current volume:** [INSERT — e.g. 40 requests per month] **Who currently handles it:** [INSERT — e.g. in-house counsel, paralegal] **Known risk factors:** [INSERT — e.g. liability cap, data processing terms, indemnities, termination rights, governing law] **Business team who would run this themselves:** [INSERT — e.g. procurement, sales operations, HR] --- ## OUTPUT FORMAT ### 1. Risk Threshold Definition Define the specific, objective criteria that separate "self-serve" matters from "escalate to counsel" matters. Express each as a yes/no or numeric test, for example: * Contract value below £[X] * No deviation from the approved clause set * Term under [Y] months * Counterparty is not on the restricted/high-risk list List every threshold needed to fully resolve this matter type without ambiguity. ### 2. Decision Tree Lay out the full triage logic as a step-by-step decision tree, starting from intake and ending in one of two outcomes: **Approved for Self-Serve** or **Escalate to Counsel**. Include every branch and edge case implied by the risk factors above. Present this as a numbered sequence of questions a non-lawyer could follow without help. ### 3. Self-Serve Intranet Page Copy Draft the actual page copy for an intranet page that walks a non-lawyer through the decision tree. Include: * A plain-language explanation of what they can and cannot approve themselves * The intake form fields needed to capture everything the triage logic requires * What happens automatically when a matter is escalated, and what the business team should expect next ### 4. Escalation Triggers & Audit Trail List the specific triggers that force escalation regardless of the decision tree's outcome (for example: counterparty objects to standard terms, unusual governing law, a relevant regulatory change). Define what record should be kept of every self-serve decision, so the process can be audited later. ### 5. Review Cadence Recommend how often the risk thresholds should be reviewed, who should own that review, and what data — volume, escalation rate, error rate — should trigger an earlier review than scheduled. ### 6. Rollout Plan Provide a practical two-to-three-week rollout plan: who to brief first, how to pilot with a small volume before full rollout, and how to measure whether the process is actually working once live. --- ## ADDITIONAL INSTRUCTION Be specific and operational, not theoretical. Write the decision tree and the intranet page copy so they are ready to use immediately, with no further editing required to make them sound human and clear to a non-lawyer. Finally, take the self-serve page from Section 3 and output it again as a complete, ready-to-upload HTML page — full markup with inline CSS, no external dependencies — so it can be published directly to a website or intranet without any further development work.

Want More Prompts & Workflows?

For prompts and more detailed workflows like this one, take a look at our book on AI for legal professionals. Explore step-by-step guides, ready-to-use prompts, and best practices for integrating AI safely into your legal work.

Explore the Book →

Until next Monday,

— Richard Nicholas

Got a story I should cover, or a question about something here? Just hit reply — I read every one. (Received this one secondhand? Reach me directly at richard@skilldiligence.com.)

Get This Newsletter Every Week

Curated AI and legal governance news, a copy-paste prompt, and a practical workshop — delivered free to in-house counsel every Monday.

Get Next Monday's Issue →