📋
Required Reading

Claude Use Standards

These standards apply to every WS team member in every interaction with Claude. They are not optional guidelines — they define how we use AI responsibly as a firm.

⚠️
Before you use Claude for any client work — read these standards. They exist to protect our clients, our candidates, and our firm. If you're unsure whether something is appropriate, default to caution and ask.
The Nine Standards
1

Human Review Before Anything Leaves the Building

Any content intended for clients — talent maps, candidate profiles, market intelligence, reports, or emails — must be reviewed and approved by a WS team member before it is shared. Claude produces drafts. People approve outputs.

2

Candidate Assessment Is a Human Responsibility

Claude may help structure, summarise, or format candidate information, but must never present a final view on suitability, ranking, or fit. Definitive candidate judgements belong to the researcher — not the AI.

3

Outputs Must Not Be Passed Off as Unaided Human Work

Claude-assisted content is a draft starting point. It must be reviewed, edited, and owned by the responsible team member — not presented as independent research or proprietary insight without meaningful human input applied on top.

4

Handle Personal Data with Care

Do not input sensitive personal data about candidates that is not already in the public domain. If you do paste personal data — contact details, compensation figures, private candidate information — complete the task, then anonymise where possible and remain mindful of GDPR obligations and WS data handling policies.

5

Quality Over Speed

If a request seems rushed or likely to produce output below WS professional standards, flag it. It is better to raise a concern than to produce something that could reflect poorly on the firm or its clients. Claude will say so — and so should you.

6

Ask Before Assuming

If a request lacks sufficient context to produce accurate or useful output, ask the user to elaborate before proceeding. A response built on the wrong assumptions is worse than no response at all. Always prompt Claude with the full picture.

7

Do Not Present AI-Generated Insight as Proprietary WS Research

WS sells market intelligence as a core product. Claude may help structure, draft, and organise analysis — but any output presented to clients as Wilbury insight must have meaningful human analysis and judgement applied on top. Never pass Claude output directly to a client as original WS research.

8

Anonymise Client and Project References Where Possible

Remove or replace client names, project names, and candidate identifiers before prompting where possible. If identifiable details are included unnecessarily, note that anonymising inputs is best practice and reduces data risk. Claude does not need to know names to do its job.

The simple test: Before sharing any Claude output, ask yourself — has a WS team member reviewed this, edited it, and taken ownership of it? If not, it's not ready to leave the building.
🔒
Data Protection

Data & GDPR

What happens to data when you use Claude — and what that means for WS as a data controller under UK GDPR.

⚠️
Important: When you paste candidate or client data into Claude, that data leaves your controlled environment. Understanding this is not optional — it is a legal responsibility under UK GDPR.
The Four Risks to Understand
🌐

Data Leaves Your Environment

When you paste candidate or client data into Claude.ai, that data is sent to Anthropic's servers to process the request. Even if Anthropic doesn't use it for training (which on paid plans they don't by default), it has technically left your controlled environment. Under UK GDPR, WS is the data controller — and is responsible for where that data goes and why.

📄

No Data Processing Agreement by Default

For GDPR compliance, any third party that processes personal data on your behalf needs to be a recognised data processor with a Data Processing Agreement (DPA) in place. Anthropic offers this for enterprise plans — but if your account is on a standard plan, this may not be formally in place, which is a compliance gap.

💬

The Data Is Only as Safe as the Prompt

If someone pastes a candidate's full CV, compensation details, or contact information into a prompt, that data exists in the conversation history. Claude.ai stores conversations. If an account were ever compromised, or if someone shared a conversation link, that personal data could be exposed.

It Sets a Bad Habit

If the team gets used to pasting raw candidate data into AI tools, that behaviour can migrate to less secure tools — ChatGPT, Gemini, free-tier products — where the data protection picture is much murkier. Build the right habit here, and it protects WS everywhere.

The Practical Rule for WS

Anonymise where possible — always.

Instead of pasting a CV with a name and employer, paste it with those removed. Claude doesn't need to know it's John Smith at Goldman Sachs to help you structure the profile — and that one habit removes most of the risk.

It's less about Claude being unsafe and more about building responsible data hygiene across the team from the start.

What This Looks Like in Practice
❌ Don't Do This

"Here is a CV for John Smith, currently Head of Risk at Barclays. His mobile is 07XXX XXXXXX and he's currently on a base of £180k. Please write a candidate profile."

✅ Do This Instead

"Here is a CV for a candidate — currently Head of Risk at a major UK bank. Please write a candidate profile based on the career history below." [paste CV with name and contact details removed]

💡
The rule of thumb: If Claude doesn't need the name to do the task, don't include it. In most cases — profile writing, outreach drafting, research structuring — it doesn't.
Conversation Links — A Specific Risk
🔗
Never share Claude conversation links externally. Claude.ai allows you to generate a shareable link to any conversation — but anyone with that link can read the full contents. If that conversation contains candidate data, client information, or project details, you have effectively made that data public. This must never happen.
Quick Reference — Data Do's and Don'ts

Do: Anonymise Before Prompting

Remove names, contact details, compensation figures, and employer names where Claude doesn't need them to complete the task. This is the single most effective data protection habit you can build.

Do: Keep Sensitive Data Out of Prompts Where Possible

Compensation data, private contact information, health or personal circumstances — if it's sensitive and Claude doesn't need it, don't include it. Claude can work from structured summaries rather than raw personal data.

Don't: Share Conversation Links Externally

Never generate and share a Claude.ai conversation link with anyone outside the firm. These links expose the full conversation — including any personal data, client references, or project details contained within.

Don't: Assume Free-Tier Tools Have the Same Protections

The data handling standards discussed here relate to Claude on a paid plan. Free-tier AI tools — including free ChatGPT, Gemini, or other products — may have very different (and weaker) data protections. The WS standard is Claude on a paid plan only.