Claude Training for UK and Irish Private Equity and Credit Firms: AI Literacy, the FCA and the EU AI Act

1. The Short Answer: Train by Role and Keep the Record

Train each role on the Claude work it actually does, using your own redacted documents, and keep a record of who learned what and when. That gives you people who use the tool well and a file you can show when someone asks how they learned.

The rules differ on each side of the Irish Sea. In the EU, Article 4 of the AI Act, rewritten in July 2026, asks providers and deployers of AI to take measures to support their staff's AI literacy, and the European Commission says no certificate is needed.

In the UK, the FCA's October 19, 2026 paper on AI in financial services describes ongoing staff training as good practice, while stating that its findings create no new requirements.

If you are choosing an AI provider in Ireland or the UK, settle the platform before the training. Check the data terms, where data is stored and what admins control; on Claude Team and Enterprise, inputs are not used to train models by default, and data is stored in the United States.

Everything here is operational guidance. Your compliance officer and counsel decide what the rules mean for your firm, and no course or tool meets a firm's obligations on its own.

2. Why Staff Training Is Now a Regulatory Question

The EU's AI literacy duty has applied since February 2, 2025. In 2026 it was rewritten and given an enforcement date, Ireland passed its own AI law, and the FCA set out what good AI practice looks like.

The Irish funds supervisor has said where it is looking. The Central Bank's 2026 Regulatory and Supervisory Outlook lists, for funds, "Continued engagement to understand usage of AI in their business models."

It adds a warning: "The growing deployment of AI across the funds sector brings its own set of risks when awareness, governance and controls are inadequate."

Awareness comes first in that list. Training is how a firm shows it has some.

3. Ireland and the EU AI Act: Article 4 After the 2026 Changes

Article 4, on AI literacy, has applied since February 2, 2025. The Digital Omnibus on AI, Regulation (EU) 2026/1744 of July 8, 2026, rewrote it; the regulation was published on July 24 and entered into force on July 27, 2026 (EUR-Lex).

Providers and deployers must now "take measures to support the development of AI literacy of their staff". The new text adds: "This obligation does not require providers or deployers to guarantee any specific level of AI literacy of any individual."

That softened the duty. It centers on the measures a firm takes, and what counts as enough for your firm is counsel's call.

The European Commission's AI literacy Q&A, last updated October 20, 2026, settles the certificate question: "There is no need for a certificate. Organisations can keep an internal record of trainings and/or other guiding initiatives."

It also sets the clock: "The national market surveillance authorities will start supervising and enforcing the rules as of 2 August 2026."

In Ireland, the Regulation of Artificial Intelligence Act 2026 (No. 31 of 2026), signed on July 21 and commenced on July 31, 2026, establishes the AI Office of Ireland, Oifig IS na hEireann. Its functions include to "promote and foster AI innovation and literacy" (Irish Statute Book).

One question stays open. The sources checked for this guide do not say which Irish body supervises Article 4 for fund managers, so ask counsel rather than assume.

The Central Bank's own expectations sit alongside. Deputy Governor McMunn said on June 8, 2026: "Firms, including fund management companies, deploying AI must be able to explain how their models work, who is accountable for their outputs, and how they are governed."

For UK firms, the Act reaches deployers outside the EU only "where the output produced by the AI system is used in the Union" (Article 2(1)(c), EUR-Lex). If your work lands with an Irish fund, an EU co-investor or an EU portfolio company, ask counsel whether that brings you in.

Most firms using Claude for documents are deployers of a general-purpose tool, and high-risk obligations for Annex III systems now apply from December 2, 2027. Counsel decides any classification; the literacy duty applies now.

4. The UK and the FCA

On October 19, 2026, the FCA published AI in financial services: good practice and areas for improvement. The paper says its findings "do not create new requirements", and it describes what the FCA saw firms doing well and where they fell short.

Four lines bear on training:

• Ongoing training. "Firms gave staff ongoing training and carried out capability reviews of their human-in-the-loop and oversight functions." Ongoing means more than a launch session.
• Real review. "There was a risk of human review of AI outputs being a 'box-ticking' exercise, rather than meaningful oversight." Teach reviewers what a real check looks like on your own documents.
• Senior managers. They need not be technical specialists, but they should "take reasonable steps to understand the risks arising from AI use within their area of responsibility".
• Outsourcing. "Firms can outsource AI services, but remain responsible for them." Buying Claude seats or Copilot licenses leaves the responsibility with your firm.

One example concerns records: "Firms kept records of consumer prompts, outputs, model versions, applied filters and escalation decisions." It describes consumer-facing firms and is an example of good practice, so treat it as an idea to borrow rather than a rule.

For partners, the line to remember is the one about senior managers. Understanding the AI risks in your area takes a good briefing and some hands-on time.

5. What Counts as Evidence

The Commission says firms can keep an internal record of trainings. A useful record holds six things:

• A curriculum by role. What partners, deal teams, credit teams, IR, finance and portfolio company staff each learned, with the materials used.
• Attendance and dates. Who attended which session, and when.
• The tools covered. The plans and settings in force at the time, such as Claude Team or Enterprise, Claude inside Copilot and the Cowork approval settings.
• Policy acknowledgment. Each person's dated acknowledgment of the AI use policy.
• Refreshers. A schedule, kept. The FCA's word is "ongoing", and the tools move: on September 16, 2026, Anthropic announced that Cowork and chat are merging into one Claude, starting with Pro and Max.
• Capability checks. A short exercise in which reviewers find planted errors in an AI draft, with the results recorded. It mirrors the FCA's capability reviews of human oversight.

Keep the file with your other compliance records and give it one owner. Then test it against Deputy Governor McMunn's line: could you explain how the models work, who is accountable for their outputs and how they are governed?

Counsel decides what your file must contain. Training's job is to make sure the file describes something real.

6. A Role-Based Claude Curriculum

Everyone starts with one session on the rules: which Claude plan the firm uses and what its terms mean, what goes in and what never does, and how to check an output.

That session also covers approvals. In Cowork on Team and Enterprise, "Automatically approve" mode is on by default and "Always allow" for connector tools is off by default, so people need to know what the firm has set and why.

Then each role works on its own material:

• Partners and the IC. What Claude does well and where it fails quietly, the data rules leadership owns, and who is accountable for an AI-assisted memo. This is the FCA's senior manager line in practice.
• Deal teams. Screening a folder of redacted CIMs against the firm's criteria, diligence questions over a working folder, and IC memo first drafts in house format.
• Credit teams. Spreading, covenants, credit memos and portfolio risk notes, each checked against the source documents.
• Investor relations. LP letters and DDQ answers drafted from approved material, with every number traced back.
• Finance and operations. Close support, variance commentary and board pack drafts, with the controller checking the figures.
• Portfolio company teams. Function cohorts (finance and accounting; sales, RevOps and marketing; operations and support) grouped by skill level, each on the company's own work.

Run the role sessions on redacted documents from your own deals, borrowers and funds. The file then shows training on the work as your firm actually does it.

The Claude training playbook covers what to teach first. For UK and Irish firms, the difference is the record you keep.

7. Data Location Questions UK and Irish Firms Ask

Settle these before the first session, because someone will ask.

• Where is our data stored? In the United States. Anthropic offers no EU storage option for Team or Enterprise.
• Is it used to train models? Not by default on commercial plans. Anthropic's Privacy Center says: "By default, we will not use your inputs or outputs from our commercial products (e.g. Claude for Work, Anthropic API, Claude Gov, etc.) to train our models." Thumbs up or down feedback is the exception, storing the whole conversation for up to 5 years, and Team and Enterprise owners can switch chat ratings off.
• How long is it kept? Deleted chats leave Anthropic's back end within 30 days on all plans, and Enterprise owners can set a custom retention period of at least 30 days. Local Cowork history stays on the user's computer, outside those settings, where admins can't delete it centrally.
• What about Claude inside Microsoft Copilot? Microsoft says Anthropic models "are currently excluded from the EU Data Boundary, and when applicable, in-country processing commitments" (Microsoft Learn). The Copilot guide covers which settings are on by default in UK and Irish tenants.

Personal accounts are the gap. Free, Pro and Max run on consumer terms, where chats can train models unless the user opts out, and Pro and Max have no organization admin console, audit logs or Compliance API. Training should say plainly that firm and portfolio company data stays off them.

Whether a transfer to the United States fits your obligations, and on what terms, is for counsel. The security guide compares the plans; your training makes sure staff know which route is approved and why.

8. Portfolio Companies in the UK and Ireland

UK Private Capital, formerly the BVCA, surveyed 76 private capital investors and reported on September 17, 2026: "Nine in ten private capital firms are encouraging the use of AI across their portfolio companies."

The detail shows a gap. Of those surveyed, 93% encourage AI for administrative and back-office tasks, while 45% provide guidance on AI governance, risk and compliance (UK Private Capital).

Training belongs in that gap. Encouragement without guidance can leave each person improvising, sometimes on a personal account.

At a UK or Irish portfolio company, the basics look like this:

• Its own Claude organization on Team or Enterprise, so company data sits under the company's own settings.
• A one-page policy covering what goes in, what never does and who approves new tools.
• Cohorts by function and skill level, up to 12 people each, on the company's own work.
• Its own training record, kept like the fund's, since the company may be a deployer in its own right; counsel decides.

If AI sits in a value creation plan, put the policy and the first cohort in the first 100 days. The portfolio company training guide sets out the cohort model.

9. Live Online, in UK and Irish Business Hours

Most of this training works well live and online. Small groups can work on their own documents with an instructor watching, and the meeting platform can record who attended.

When you compare providers, ask each one the same questions:

1. Will sessions run in our business hours? London and Dublin keep the same clock, so one schedule covers both offices.
2. Is it hands-on, on our redacted documents? A demonstration on someone else's examples leaves little behind.
3. What do we get for the file? A curriculum by role, attendance records and dates, in a form compliance can keep.
4. How big are the groups? Up to 12 people lets everyone work on their own material.
5. What happens afterward? Refreshers, office hours and a plan for the first 30 days.
6. Is the provider independent of the licenses? A trainer paid a referral fee on seats has a reason to sell more of them.

Compare the answers on paper. The file you want at the end should shape the provider you pick at the start.

10. What It Costs

Training has two costs, the fee and the team's time, so budget both.

The time is easy to estimate. Assume a 12-person deal team in three 2-hour sessions: that comes to 72 person-hours, before any preparation.

Licenses are separate from any training fee. Anthropic's pricing page lists Team at $25 per standard seat a month, or $20 billed annually, and Enterprise at $20 per seat per month billed annually plus all usage at API rates, with a 20-seat minimum when bought self-serve.

Skipping training has a cost too: seats nobody uses and a training file with nothing in it.

11. Where to Start

If you do one thing this month, write down which Claude plan and settings your firm uses and who has been trained on them. That page is the start of your file.

1. Read the rewritten Article 4 and the FCA paper with counsel, and agree what your file needs to show.
2. Settle the platform and its data terms, including whether anyone can pick Claude inside Copilot.
3. Write the one-page policy and collect acknowledgments.
4. Brief the partners first, then run role sessions on your own documents.
5. Start the file: curriculum, attendance, dates, and the tools and settings covered.
6. Book the refreshers and the first capability check for reviewers.
7. Extend it to portfolio companies in the UK and Ireland, each with its own record.

The EU AI Act checklist covers the rest of the Act for portfolio companies. Training is the part you can start this month.