Your managers are training your competitors' AI with your performance reviews. Stop them this week.

I've sat in three CHRO offices in the last six months where the same scene played out: the legal team realizing that managers across the org had been pasting unredacted performance reviews — names, comp, ratings, the works — into the free tier of ChatGPT to 'help write better feedback.' That data is, depending on the tier, in the training corpus or in retained chat logs that the vendor can subpoena, get breached, or change policy on tomorrow.

The 5-step lockdown (one week, no budget required)

• Day 1: send a one-paragraph policy email naming the specific tools that are and are not approved for HR content. Be concrete, not aspirational.
• Day 2–3: deploy DLP rules in your email and browser that flag pasting of names + ratings + comp into non-approved domains. Most enterprise security suites already have a template.
• Day 3: stand up an approved alternative — either an enterprise tenant of Copilot/ChatGPT with zero-retention and no-training, or a self-hosted model. The reason people use the consumer tool is that it's the only one available.
• Day 4: rewrite your manager enablement: 'AI can draft, but only inside the approved tool, and only with PII removed.' Show, don't tell — give them three before/after examples.
• Day 5: add a one-line attestation in your performance review tool: 'I have not pasted any portion of this review into a non-approved AI tool.' This isn't legal protection; it's a behavior nudge that works.