According to Politico, Capitol Hill is introducing a draft bill behind the anthropomorphic veil to protect the most advanced AI models. It’s certainly not often that an AI lab tries to write its own industry standard. But what has raised more than one eyebrow is that his proposal would set the bar too high before the government could stop a model, concentrating that power in a single department. We tell you the details.
work in progress Congress in the US is currently weighing several AI protection proposals, and Anthropic has drafted two or three notable ones. An Anthropology spokesperson told Politico that the bill reflects the company’s Advanced AI Framework, which it has been promoting to lawmakers since June, including its content as “sound policy that we believe Congress should be working on.”
Andl Threshold. According to Politico, which has access to the draft, the government can only stop the launch or use of a model if it poses a potential and imminent risk of more than $1 billion in damages, or the death or serious injury of more than 50 people. Danger must also come from:
- A biological weapon.
- A model that escapes the control of its developer.
- cyber attack
How will it work. Media reports that the independent institute, registered with the Department of Commerce, will periodically evaluate an AI laboratory and its frontier models. Laboratories may pay them, as long as the payment does not depend on the results of the evaluation. If an appraiser identifies a “listed imminent risk,” he or she must notify the Department of Commerce within 72 hours of completing the appraisal. After that, the trade could ask a federal district court judge to stop launching a model, according to media reports. Tom’s Hardware adds that the trade must present its evidence to a judge, who will make the final decision.
orOnly one organism has a switch. Politico says the text expressly prohibits any other “department, agency or official” from suspending the development, deployment or internal use of a model. So if we combine that with the threshold explained earlier, it makes the Commerce Department the only executive agency to shut down a model, and only a court can order it.
Regardless of who owns that red button, Politico says AI labs must create security frameworks, publish transparency reports for new frontier models, protect those who report model security issues, and report serious security incidents to the Commerce Department.
A delicate moment. Concerns about AI security have intensified after a number of notable incidents in the past few months, including hacking Face Hacking by hundreds of OpenAI agents, or the break-in of several AI agents on government websites, among other critical moments.
Politico added that several incidents that emerged this summer did not result in more than $1 billion in fatalities or economic damage, so from the perspective of the draft shared by Anthropic, those incidents would not cross that threshold.
It’s only a draft at this point, and Anthropic has yet to formally approve any specific bills in Congress. We’ll have to wait and see if lawmakers see $1 billion and 50 victims as a reasonable security measure. It is doubtful that the concept has stirred up the sector at all.
In Xataka | TikTok owner Zhang Yiming doubled his fortune. Their secret: AI, of course
