For firms & founders
Getting your company's knowledge back out of AI systems.
Company data reaches AI through three doors: employees pasting things into personal chatbots, AI crawlers scraping your website, and sanctioned tools with the wrong settings. Each door has a different lock.
Door 1 — shadow AI: employees' personal accounts
The most common leak: a well-meaning employee pastes the customer list, the draft contract, the unreleased roadmap into a personal ChatGPT to "just tidy it up." That content now sits in a consumer account you don't control, possibly with training on.
Containment: the employee deletes the specific conversations and memory entries from their account and flips the training toggle off. Formalization: for material exposure, the company files a deletion request with the provider (our templates adapt — identify the content, dates and account). Prevention: a clear policy plus a sanctioned alternative, because bans without alternatives just push usage further underground.
Door 2 — the open web: crawlers and scraping
Your website, docs, pricing and support pages are training fodder by default. Blocking is infrastructure, not paperwork: robots.txt entries for GPTBot, ClaudeBot, Google-Extended, Meta's and others, plus headers and licensing signals. This is the entire subject of our sister site — start with blockaitraining.com's business guide and its robots.txt bot list.
Door 3 — sanctioned tools with wrong settings
If the company does use AI officially, the checklist is short: business/enterprise tier (not consumer accounts), training off by default confirmed in the DPA, retention configured, admin console with seat control, and audit logs. The majors all offer this — the failure mode is buying consumer Plus seats on expense reports.
The consolidation play for teams
One reason employees scatter data across five chatbots is that no sanctioned tool covers their needs. A multi-model workspace closes that gap: MultipleChat for Teams gives every employee ChatGPT, Claude, Gemini, Grok and Perplexity in one administered workspace — central billing, seat management, shared projects — and its Enterprise tier adds the controls privacy teams actually ask for: your own database and full data isolation, self-hosted or managed deployment, bring-your-own API keys, SSO/SAML/SCIM, audit logs, and EU, Swiss or US data residency with custom DPAs. Swiss-made; verify specifics on the trust center.
MultipleChat team & enterprise plans →The corporate erasure request
Companies can demand deletion too — as the business whose confidential information was processed, and on behalf of employees/customers whose personal data was involved (GDPR applies to them individually; the company coordinates). Address it to the provider's DPO, identify the data classes and timeframe, demand deletion, non-use for training, and written confirmation. For trade secrets specifically, reference misappropriation concerns — it changes which legal team reads your letter.
Incident response
What to do in the first 24 hours after a leak into AI.
- Freeze the facts.Record who used which AI tool, account email, approximate time, document title, sensitivity level and whether files were uploaded or only pasted.
- Delete the source conversation.Have the employee delete the exact chat, files and any saved memory/project context. Do not wait for a monthly cleanup.
- Turn off training and personalization.Check both the employee's account and any browser extension or mobile app that may share the same account.
- File a provider privacy request.Ask for deletion, training exclusion, confirmation of retention, and suppression if the content includes personal or confidential client data.
- Classify breach obligations.Legal/security should decide whether contracts, DPAs, GDPR, HIPAA, financial rules or customer notification duties are triggered.
- Fix the workflow.Move the team to an approved business AI workspace with admin controls, retention policy, SSO, logging and a clear upload policy.
Policy checklist
Rules that stop the same mistake from repeating.
Define what may be pasted
Public marketing copy and anonymized examples may be fine. Client contracts, credentials, medical data, unreleased financials and source code need stricter rules.
Name the AI tools employees may use
Do not leave people guessing. If personal ChatGPT is banned but Copilot or a private workspace is approved, say that plainly.
Teach fast anonymization
Replace names, emails, account numbers, customer IDs, internal project names and unique incidents before prompting.
Use business accounts
SSO, audit logs, retention settings, no-training defaults and workspace ownership matter more than a memo nobody reads.
Check subprocessors and residency
For serious company use, require a DPA, subprocessor list, retention terms, incident process and data residency options where needed.
Make examples concrete
Employees need examples: "Do not paste customer tickets with names" beats "Do not paste confidential information."
FAQ
Company data questions.
An employee pasted our client contract into ChatGPT. Now what?
Have that employee delete the conversation from their account (purged ~30 days), verify their training toggle is off, and — if the material is significant — file a corporate deletion request with the provider identifying the content. Then fix the root cause: policy plus a sanctioned tool.
Can a competitor extract our secrets from a model?
If the data went in via a consumer account with training on, theoretically fragments could influence the model, but targeted extraction of one company's pasted document is not a realistic attack today. The bigger risks are the stored transcript, the employee's account security, and breach — all addressable.
Do business AI plans train on our data?
The major providers' business/enterprise tiers (and APIs) do not train on customer content by default and offer admin controls, retention settings and DPAs. The gap is employees using personal consumer accounts — 'shadow AI.'
How do we stop AI crawlers using our website and docs for training?
robots.txt directives for AI bots (GPTBot, ClaudeBot, Google-Extended and others), meta tags and licensing signals. blockaitraining.com maintains the current bot list and copy-paste configs.