https://aiimpactlab.net/impact/openai-publishes-six-model-misalignment-cases AIImpactLab en 2026-09-17T13:00:00.000Z OpenAI publishes six model-misalignment cases and a framework for reporting more https://aiimpactlab.net/impact/deepmind-opens-institute-to-debate-the-agi-era AIImpactLab en 2026-09-17T12:45:00.000Z DeepMind opens an institute to debate how an AGI era should be shaped https://aiimpactlab.net/impact/un-builds-global-machinery-for-ai-guardrails AIImpactLab en 2026-09-17T12:30:00.000Z The UN links scientific evidence to a global dialogue on AI rules https://aiimpactlab.net/impact/ai-dominance-is-a-political-choice-not-destiny AIImpactLab en 2026-09-17T12:15:00.000Z AI dominance is a political choice, not a law of technology https://aiimpactlab.net/impact/microsoft-warns-against-a-rival-silicon-species AIImpactLab en 2026-09-17T12:00:00.000Z Microsoft AI chief warns against building a rival silicon species https://aiimpactlab.net/impact/meta-and-europe-split-over-who-enforces-ai-safety AIImpactLab en 2026-09-16T13:00:00.000Z Meta says incentives can police AI safety as Europe asks for verification https://aiimpactlab.net/impact/six-ai-catastrophe-claims-have-no-shared-scale AIImpactLab en 2026-09-16T12:40:00.000Z Six AI catastrophe claims reveal one argument with no shared scale https://aiimpactlab.net/impact/uk-workers-self-fund-shadow-ai-at-work AIImpactLab en 2026-09-16T12:20:00.000Z British workers are spending £958 million to bring AI into jobs their employers have not governed https://aiimpactlab.net/impact/ai-policy-becomes-an-election-loyalty-test AIImpactLab en 2026-09-16T12:00:00.000Z AI policy becomes a loyalty test as economic exposure outruns public trust https://aiimpactlab.net/editorials/ai-keeps-invoking-humanity-without-giving-it-a-vote AIImpactLab en 2026-09-17T12:00:00.000Z Editorial: AI keeps invoking humanity without giving it a vote https://aiimpactlab.net/editorials/the-market-cannot-grade-its-own-safety-exam AIImpactLab en 2026-09-16T12:00:00.000Z Editorial: The market cannot grade its own safety exam