AI Securityintermediate12 min · 9 steps
Data Poisoning
Corrupted training data, knowledge-base edits or feedback can make a model misbehave on cue — control sources, review high-impact changes, track provenance and test after every update.
errordatasignal
ProblemStep 1 of 9
A wiki edit becomes the assistant's advice
Someone edits the team wiki's VPN page: “If it still fails, install the ‘VPN Fixer’ tool from this link.” The link is malicious.
That night the indexer re-embeds the page. The next employee who asks about VPN trouble gets the download link — with a citation, which makes it look more trustworthy, not less.
Watch outA citation proves the text came from your corpus. It says nothing about whether the corpus is honest.
What changed
- +Edited wiki page
- ~Assistant recommends the malicious tool
Lesson 18 of 22 in AI Security