Misinformation

LLMs generate fluent, plausible text that can be factually wrong (hallucination). When applications present this output as authoritative — or agents act on it — the result is misinformation with real consequences, from bad code to unsafe advice.

How it shows up

• A coding assistant invents a non-existent package name (leading to slopsquatting risk).
• A support bot confidently states an incorrect policy.
• A model fabricates citations that do not exist.

Mitigations

• Ground responses with retrieval and require citations for consequential claims.
• Add human review for high-stakes decisions.
• Communicate uncertainty and limitations to users.
• Validate generated artifacts (e.g. verify packages and links exist).

Related threats