AI firms disclose 'autonomy' thresholds

Companies are beginning to state, in filings or public documents, quantitative or descriptive thresholds for when they would consider models 'autonomously' accelerating development (for example, a claim that future models may meet a threshold of “dramatically accelerated progress attributable to AI”). Making such thresholds explicit turns an internal safety/governance standard into a public metric that investors, regulators, and civil society can respond to. — If firms codify what counts as self‑improvement, regulators and investors gain a concrete lever to judge, audit, and demand safeguards around autonomous model development.