MIT study: 32% of top 100 companies publicly acknowledge zero AI risks
rohanpaul_ai · x · 2026-11-05
An MIT FutureTech / University of Queensland paper (arXiv:2611.04278) coded public documents from the world's largest companies against the 24 risks in the MIT AI Risk Repository, benchmarked against a Delphi study of 272 AI risk experts.
Key findings:
- 32% of the top 100 companies acknowledged no AI risks; an estimated 62% of the top 1000 were silent.
- Five risks (discrimination, weapons & cyberattacks, fraud, privacy loss, devaluation of human creativity) account for over half of all acknowledgements.
- The most severe risks are almost absent: no top-100 company mentioned power centralization or competitive AI racing; only 4 mentioned dangerous capabilities.
- General-purpose AI developers named misalignment only 15% of the time vs 85% for discrimination.
- Companies in more vulnerable sectors disclose more, but the most responsible roles don't disclose more than others.
The authors argue companies should name each risk alongside its safeguard, providing an empirical baseline for AI risk disclosure requirements.
Related event: MIT Study: Nearly a Third of Top Global Companies Disclose No AI Risks(2 posts)→
More from Companies & People
- Fireworks AI cofounder: training is now an economics play for AI apps — sophiamyang · 2026-11-05
- Joke about OpenAI org chart: "everyone reports to everyone else" — a dense graph — angadsg · 2026-11-05
- fal marks new funding round with an inside look at its co-founders' partnership — noahsolomon · 2026-11-05
- Amazon MGM explains dropping Sam Altman biopic: film didn't match its ethos, Bezos not involved — KateClarkTweets · 2026-11-05
- OpenAI's Dean Ball names GPT-4, Opus 3/4.5, o3 and Astra among his favorite tech releases ever — deanwball · 2026-11-05
- Apple Reportedly Slashes Internal AI Budgets for Engineering Teams — TechCrunch AI · 2026-11-05