Misinformation risk arises when confident but inaccurate model output is presented as reliable fact, leading to poor decisions or unsafe actions. Attackers can intentionally steer outputs through adversarial context and deceptive prompts. Without grounding and verification, misinformation can propagate quickly through automated workflows.
Ground responses in trusted sources, require citation-aware workflows, and implement verification checkpoints for high-impact content. Display uncertainty indicators and avoid presenting probabilistic output as deterministic truth. Monitor and evaluate misinformation patterns continuously to improve safeguards.
The same feature written the risky way, then with the fix applied.
How this risk has ranked in each OWASP release.
2025Latest
#09llm09-2025 Misinformation
