Data Sanitizer · Blog

How to redact logs before pasting them into AI

Log lines are one of the most common things people paste into AI — and one of the leakiest. A single stack trace can expose your internal network, a service account and a live API key in three lines.

What leaks in a typical log line

Take an ordinary auth event:

account=CONTOSO\jsmith src=10.0.0.143 host=rkfd-nu-sql api_key=sk-live-4f8aE2bQ

That one line exposes a domain and username, an internal IP, a server name that hints at role and location, and a live secret. Multiply that across a paste of 200 lines and you've handed over a map of your environment.

What to strip

• Secrets: API keys (AWS, Google, GitHub, Stripe, OpenAI), JWTs, bearer tokens, password= values, private keys.
• Network: internal IPv4/IPv6, hostnames, FQDNs, cloud ARNs and resource IDs.
• Identity: domain\user, emails, SIDs, GUIDs, MAC addresses.

Keep it useful for the AI

Don't just blank values out — that destroys the AI's ability to correlate events. Use deterministic tokens: every occurrence of the same value maps to the same placeholder, so 10.0.0.143 is always IP_INT_01. The model still sees "this IP talked to that host 40 times" — it just never sees the real address.

Do it in one step

Manually find-and-replacing is slow and error-prone — you will miss one. An automatic, pattern-based redactor catches the whole class of values in a click, shows you exactly what it changed, and lets you restore the AI's answer to real values afterward. And because it runs in your browser, the raw log never leaves your machine.