Framing bias is a subtle form of bias where the way an answer is presented shapes how people interpret it. In AI systems, the same facts can be delivered with different emphasis, wording, or order, causing a response to sound neutral while still nudging the user toward a particular conclusion.
This matters in cyber security because analysts, managers, and the public often use AI-generated summaries, incident triage, policy drafts, and support responses. Framing bias can make a risky event seem minor, present one mitigation as the only sensible option, or downplay competing explanations. Defenders should test models for both factual accuracy and presentation bias, especially in high-impact workflows. Attackers can also exploit framing by crafting prompts or content that steer a model toward misleading interpretations.