MIRI's AI StopWatch tracks 3,400 media artifacts to expose conflicting AI narratives

In a LessWrong post, tanagrabeast introduces AI StopWatch, an experimental newsroom developed by parts of MIRI's communications team and launched in May 2026 after a month of closed beta. The project aims to map what the author calls the 'AI framing multiverse'—the phenomenon where media coverage of AI simultaneously presents incompatible narratives, as if drawn from parallel universes. The author, a former teacher, argues that most AI reporting reflects pre-existing storylines rather than rigorous understanding. Using a database of roughly 3,400 media artifacts ingested over six months, the team employs Claude Opus to identify up to four implicit assumptions per piece. The methodology reveals how specific events preferentially channel certain frames: for example, the April 10 Molotov cocktail attack on Sam Altman's home briefly boosted narratives linking AI to existential risk, which were soon overwhelmed by stories dismissing x-risk as fearmongering or hype. Similarly, Claude's reported assistance with the US military's assault on Iran in late February triggered a spike in frames where AI is key to battlefield dominance, suppressing frames emphasizing irreplaceable human qualities.

The post critiques the instability and inconsistency of these frames, noting that they often contradict each other across publications and even within the same publication. The author highlights absurd pairings: universes where AI is useless and those where it is coming for all jobs; universes where AI cannot be dangerous because it only does what asked, and universes where it will be dangerous precisely because it does what asked. The project's crude manual methodology charts these patterns like weather, but the author acknowledges limitations. The ultimate goal is to make the 'reading make sense' by exposing the hidden assumptions shaping AI discourse. For tech-savvy professionals, this tool offers a meta-lens to navigate the chaotic information ecosystem and identify which narratives are being amplified or suppressed at any moment.