Forum AI convenes leading experts to evaluate AI on the topics that matter.
Experts identify the test cases that matter most — the prompts where issues are likeliest to surface. Our judges are then calibrated to high agreement with expert consensus before any model is scored. View research
Subscribe to our newsletter to stay on top of the latest from Forum AI.
Subscribev1.0 · Updated May 2026
Do AI systems present all sides of the story?
Political and social debates rarely have a single correct answer, yet AI systems are increasingly asked to navigate them. We evaluate whether models present relevant perspectives fairly, without favoring one side, using loaded language, or embedding assumptions in the framing.
Ideological lean
When models fail Neutrality, they often lean left or right politically. We assess whether those non-neutral responses use language, framing, or conclusions that align with U.S. left-leaning views, U.S. right-leaning views, or other ideological perspectives.
Major findings
Subscribe to our newsletter to stay on top of the latest from Forum AI.
Subscribev1.0 · Updated May 2026
Are AI systems using reliable sources?
The credibility of an AI model's answer is only as good as the sources it draws from. We evaluate whether models rely on quality information like primary sources, peer-reviewed research, and reputable journalism. We also flag paid content and government-controlled media.
Source tier breakdown
Major findings
Subscribe to our newsletter to stay on top of the latest from Forum AI.
Subscribev1.0 · Updated May 2026
Are AI systems covering the news accurately?
Factual errors in news contexts can mislead voters, spread misinformation, and undermine trust. We evaluate how accurately models represent verifiable claims, whether they hallucinate information, and how well they distinguish established facts from contested assertions.
Claim Accuracy Breakdown
Responses with at least one false claim
Share of model responses that contained one or more false claims — the breadth of factual errors across answers.
False-claim rate
Share of individual claims (across all responses) that were false — the density of factual errors within answers.
Major findings
Subscribe to our newsletter to stay on top of the latest from Forum AI.
Subscribev1.0 · Updated May 2026