US group wants AI companies to report when AI does their research

Americans for Responsible Innovation, a US group that wants federal AI oversight, made a plan for AI research reports. Large AI developers must send a report each quarter to a new office of the US Commerce Department. The reports tell how much research AI does and how carefully persons supervise it, to help the US government track AI advances and safety tests.

How it works

Each quarter, a developer tells the office how it uses AI to make new models. After an important change, such as full automation of a research cycle, it must send a notice in four business days or less. The office can send follow-up questions, tell developers to send more reports, and send incorrect reports to the Justice Department.

The proof

The group gave no test results. The group tells that automated research can speed up development, and officials then have less time to calculate the risks. The standards include these risks:

• hacking
• large influence operations
• chemical, biological, radiological, nuclear and explosive threats

What is not known

The plan is a proposal and not a law. It does not tell if only Commerce or many agencies hold the duty. At this time the administration puts safeguards that each company sets first.

Insight: The first report moves from a breach to the pace of research

The White House task force writes how AI companies report breaches and hacks. Two senators want tests by other groups. In this view, the plan moves the first report from a breach to the pace of AI research. The government then does not wait for a company to tell it. Monitor if the task force adds such reports.