One resignation turned the embers of AI fear into a wildfire

Details

The resignation of AI researcher Jacob Coxon, citing safety concerns, ignited a wider public debate about existential AI risk than anyone anticipated. Nathan Lambert argues in his Interconnects newsletter that this was not an accident — the AI discourse environment has fundamentally changed in 2026, as a series of high-profile breakthroughs dried out what was previously 'damp ground' that extinguished earlier safety concerns. Lambert points to two specific catalysts: the OpenAI-HuggingFace incident and OpenAI's Navier-Stokes simulation breakthrough, which together pushed more people outside the AI industry to ask whether they should be paying attention. Into this charged environment, Coxon's departure spread Evan Hubinger's >10% human extinction risk estimate far beyond the usual AI safety community. Lambert draws a sharp distinction between credible AI risks — cyber attacks on critical infrastructure, biosecurity — and the extreme extinction probability estimates he considers too low to meaningfully discuss. His concern is that public discourse conflates these under a single 'existential risk' umbrella, making productive conversation harder. For anyone tracking how AI safety narratives evolve in public consciousness — a domain increasingly shaping regulatory and corporate decisions — this piece is essential context.