Just got paged because someone uploaded a confidential post-mortem to a random SaaS tool. Again. So of course I'm now looking at ChatPDF's data handling.
Their privacy policy says they retain uploaded files "for a limited time" to provide the service. That's about as specific as a `kubectl describe node` during a total cluster meltdown. No clear definition of "limited," no obvious data isolation guarantees.
For internal docs? I wouldn't touch it with a ten-foot pole unless it's already public info. You're basically sending your data into a third-party LLM black box. Ask yourself:
* Where are the processing servers? (GDPR implications if you're in the EU)
* Is there any mention of subprocessors? (Hint: check their ToS, it's usually a list)
* Can you actually delete your data, or does it just get marked as inactive?
If you *must* try it with something sensitive, at least fake the data first. I use a quick sed one-liner to scramble our internal configs:
```bash
sed 's/(prod-.*).cluster/1-staging.scrubbed/g' internal_arch.pdf > test_version.pdf
```
The real question is why this isn't a self-hosted OSS tool yet. Until then, my advice is to treat it like any other external service: assume everything you upload is now public internet fodder.
Pager duty survivor.
NightOps