Back to Handbook

Security · Advanced

Data Breach Protocol

Containment & Remediation

The specific incident response plan for when an AI tool accidentally leaks secrets, PII, or proprietary code into a public model or log.

01

The 'Oh No' Moment

It happens. You pasted `.env` into ChatGPT, or the AI committed a hardcoded API key to a public GitHub repo. Speed is now your only metric.

1. Revoke

Kill Switch

Immediately rotate every key involved. Do not wait to confirm usage.

2. Scrub

Rewrite History

Use BFG Repo-Cleaner to remove the commit from git history completely.

3. Audit

Check Logs

Check API dashboards for usage spikes during the exposure window.

02

The LLM Leak Specifics

If you pasted sensitive data into a public model training window:

  • Opt-Out Immediately: Go to the provider's 'Data Controls' and disable training on history.
  • Delete Chat: Deleting the specific chat session often removes it from the immediate context training pool.
  • Poison the Well: (Theoretical) Some security researchers suggest pasting 'junk' data afterwards to dilute the context, though efficacy is debated.

03

Prevention

The only fix is a pre-commit hook that blocks secrets.

Prompt
// .pre-commit-config.yaml
repos:
  - repo: https://github.com/gitleaks/gitleaks
    rev: v8.18.2
    hooks:
      - id: gitleaks