What happened
A new MIT Sloan School of Management and MIT FutureTech working paper coded over 500 public documents (annual reports and AI policies) from 100 of the world's largest companies by market capitalization plus 100 randomly selected from those ranked 101-1,000 against the 24 AI risk categories in the MIT AI Risk Repository, then compared acknowledgements with a Delphi study of 272 AI risk experts. The headline finding: '32% of the world's 100 largest companies and 62% of the 1,000 largest companies have not publicly acknowledged any AI risks.' Five risk categories (discrimination, weapons and cyberattacks, fraud and scams, loss of privacy, and devaluation of human creativity) accounted for over half of all acknowledgements, while no top-100 company acknowledged power centralization or competitive dynamics, only 2% acknowledged AI misalignment, 1% multi-agent risks, and just 16% acknowledged AI security vulnerabilities. The researchers propose regulators and investors set AI-risk reporting expectations modeled on climate reporting, requiring companies to report against a standard set of AI risks plus sector- and role-matched risks and the mitigations they employ. (I verified the finding's depth on MIT Sloan's announcement page; the working paper landing page/PDF was not directly discoverable, so this is cited to the institution's announcement and sorts to the watch list.)
Why it matters
This quantifies the corporate AI-risk disclosure gap in a way that directly informs investor, audit-committee, and regulator expectations of what 'adequate' AI risk reporting should look like.
Action needed
Compare your own AI risk disclosures against the MIT AI Risk Repository's 24 categories — especially misalignment, multi-agent and dangerous capabilities — to pre-empt scrutiny and position for likely climate-style AI reporting standards.