Proposal would make AI companies reveal if AI helps build new AI

A policy group wants the biggest AI companies to tell the government whether they use their own AI to build the next version, and how much. The proposal comes from Americans for Responsible Innovation (ARI), a nonprofit of academics, former congressional staffers and government officials. Nextgov/FCW first reported it.

What the proposal says

Major developers would send quarterly reports to a new Commerce Department office. They would describe how much research work AI does and how closely people supervise it. A significant change in research practices would require notice within four business days. Fully automating a research cycle is one example that could trigger that notice.

Why anyone is asking

Building AI involves writing software, preparing training data, running experiments and deciding which results are worth pursuing. If AI takes over more of that work, new systems could arrive faster. That would leave officials and safety researchers less time to check for risks. ARI wants reporting in place before a security incident exposes a problem. In its words, Washington's visibility into that shift should not depend on what companies choose to announce.

Teeth and extras

• Commerce could ask for more information and require more frequent reports.
• Knowingly false or incomplete filings could be referred to the Justice Department.
• A whistleblower program would give insiders another way to raise concerns.
• Safety standards would cover hacking, large-scale influence operations, and chemical, biological, radiological, nuclear and explosive threats.

Where it fits in Washington

The administration has favored company-led safeguards over a broad new regulatory regime. A September industry accord calls for internal controls, outside audits and independent oversight committees. A task force led by Director of National Intelligence Jay Clayton is working to put it in place. It has also asked Pentagon technology official Emil Michael to develop protocols for labs to report breaches and hacks.

ARI's idea goes further. Incident reports cover failures that already happened. These disclosures would apply even when nothing has gone wrong.

What is not known yet

This is a proposal from an outside group, not a rule. It is not clear whether the administration will embrace it. It also leaves open whether oversight would stay within Commerce or be split among agencies. If adopted, AI firms would have to document AI's role in their research, which could matter to the vendors you rely on.