Ignacio Henriquez

**Atlas Live — Evidence-driven analysis of video claims** Atlas Live is an AI-powered newsroom that analyzes video content, extracts verifiable claims, investigates them through specialized agents, and produces evidence-backed assessments with traceable sources. The problem it addresses is the difficulty of verifying claims in fast-moving video content. Journalists and viewers often have to distinguish between what a speaker actually said, what can be independently supported, and what remains uncertain — while dealing with fragmented sources, misleading context, and limited time. Atlas Live combines transcription, claim extraction, multi-agent research, adversarial analysis, and deterministic verification into a single workflow. Specialized agents investigate claims from different angles, challenge preliminary findings, and gather supporting or contradicting evidence. A deterministic verification layer evaluates the collected evidence and produces structured results rather than relying on an LLM's unsupported judgment. Every assessment is designed to be traceable: users can inspect the underlying evidence, sources, and reasoning behind the result. The system can also abstain when the available evidence is insufficient. Atlas Live is not designed to declare an absolute truth. Its goal is to make the verification process more transparent, reproducible, and useful by separating model-generated hypotheses from evidence and deterministic evaluation. Built with Meta's Muse models, Atlas Live explores how coordinated AI agents can support more accountable video journalism.