Two academics propose that the US start tracking AI’s harms consistently before deciding how tightly to regulate the technology. Their suggested first step is a shared reporting system, with duties scaled to the risks and reach of each AI deployment.
Watch Desk analysis
What happened
In a Wharton article, Burcu Esmer and N. Bugra Ozel argue that the US should pass an AI Measurement and Incident Reporting Act. They propose directing NIST’s Center for AI Standards and Innovation to publish common definitions and reporting forms within a year.
Under their proposal, frontier developers above a capability threshold would file standardised pre-release evaluations and serious-incident reports. Large hospitals, employers and lenders using AI would report consequential errors, human overrides and outcome disparities to their existing regulators. Independent auditors would verify reports, with aggregate results made public. This is the authors’ proposal, not an announced government plan or a bill described as already introduced.
Why it matters
The authors’ central point is that harms are hard to compare when companies and institutions use different definitions, tests and reporting practices. They cite the FBI’s 2025 internet-crime report, which recorded 22,364 complaints and nearly $893 million in reported losses where AI was mentioned. They stress that those figures do not show AI caused the losses. Less visible outcomes, such as a rejected job or loan application, have no comparable national ledger in their account.
A common reporting system could give regulators and the public better evidence about what is happening across deployments. It would also raise practical questions: what counts as a serious incident, who verifies the reports, and how do you compare systems without treating unlike cases as the same? A ledger is less dramatic than a grand new AI agency, but potentially more useful than arguing over harms nobody has measured consistently.
Our read
This is a concrete policy argument: count and compare outcomes before reaching for broader constraints. The authors would put reporting duties on larger or higher-risk uses, rather than sweeping every startup and research lab into the same net. Whether their proposed thresholds and reporting rules would work is open to debate, but the focus on comparable evidence deserves one.
What to watch
- Whether US lawmakers take up a reporting framework or assign the work to existing agencies.
- How any proposed thresholds define which developers and deployments must report.
- Whether reporting would cover quieter outcomes, not only incidents that are already easy to count. Activity teaser: Two Wharton academics argue that US AI policy should start with a common way to record and compare harms. Their proposal would require reports from larger developers and from hospitals, employers and lenders using AI, with independent audits and public aggregate results. It is a policy proposal, not a bill already moving through Congress. The useful argument is whether better measurement should come before broader restrictions, and who should set the rules.
Discussion spark: Should AI developers and large organisations using AI have to report standardised harm data before regulators impose broader limits, or would that leave people exposed while the evidence is gathered?
Sources and evidence
- Source update (10 November 2026, 17:34 UTC)
Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.