Invest2 publishers3 min readPublished
Sanders attaches a 20-year prison penalty to his superintelligence ban
The bill would freeze the most advanced AI development until a new Cabinet-level Department of AI writes safety rules, then require federal approval before deployment. Researchers at OpenAI and Google DeepMind endorsed it.
The Investor · Invest desk
What happened
- Sanders and Casar unveiled legislation on Wednesday that would permanently prohibit artificial superintelligence and create a federal agency to oversee advanced AI.
- The bill would also ban recursive self-improvement and stand up the Department of Artificial Intelligence as a Cabinet-level body, pausing advanced development until it writes guidelines.
- Advanced systems would need federal approval before deployment, and violations could carry penalties including up to 20 years in prison in some cases.
- Current employees at AI companies and outside experts signed a statement backing the bill, shared with the Associated Press.
Compiled by The InvestorSomething wrong?How this is made
Why it matters
- constraint A pre-deployment approval gate puts frontier release timing inside an agency that does not exist, and the development pause has no end date until that agency is staffed and its rules are written.
- exposure Prison terms reach individuals. A levy can be modelled into a cost of capital, but no model absorbs a prison sentence, and the researchers endorsing the bill hold the jobs it would police.
- decision Winning a chamber this fall would force Democrats to choose between taxing the sector and halting its most advanced work, starting from Casar's instruction to discard the smaller bills.
- precedent Governors are moving without waiting for Congress, so the oversight a lab has to budget for arrives state by state before any federal department opens its doors.
Nothing in this bill becomes a cost to a lab this year. Semafor reported that Congress is unlikely to act on AI in the lame-duck session [19]. The Associated Press account carried by Fortune puts the bill at long odds in a Republican-controlled Congress that has not agreed on far narrower AI measures [9]. Trump has resisted recent calls for greater federal oversight [10]. Sanders told the AP, "It doesn't take a genius to say, 'slow it down.'" [8]
What a shareholder can price is the next Congress. If Democrats take the House or Senate this fall they will have to untangle a set of competing AI proposals [19], and Casar, who chairs the Congressional Progressive Caucus [11], has already set the terms of that fight. "We need to throw out these proposals that would not actually address the scale of the problems," he said [18].
Two of the three federal proposals in these two accounts take cash off the sector's balance sheet. Sanders introduced legislation in June to create a sovereign wealth fund financed by a one-time tax on the stock of the largest AI companies [12]. Sen. Mark Kelly plans a bill on Thursday creating a fund that tech companies would pay into to support workforce development [20]. Neither account specifies a rate for the tax or a size for the fund [22]. The third proposal takes the product off the market. It would permanently prohibit systems exceeding human cognitive ability, ban recursive self-improvement, and require federal approval before any advanced system is deployed [3][16][4].
Violations could carry up to 20 years in prison in some cases [4]. Criminal exposure attaches to people, and the people endorsing this bill work inside the labs it would police [5]. Juan Felipe Ceron Uribe, a researcher in OpenAI's Safety Systems, said "superintelligence could either go extremely right or extremely wrong" and that "we shouldn't be playing such games" [6]. Swante Scholz is a software engineer at Google DeepMind and said he was not speaking on behalf of his employer. On the current path of development, he said, the "most likely outcome is an existential catastrophe for humanity" [7].
The polling in the AP account is about power and water. Concern about AI's environmental impacts reached 53% extremely or very concerned, up from 41% a year earlier [15]. That is 12 points in a year, a relative rise of just under 30% [21], and the AP attributes much of the increase to Democrats [15].
In my view the levy is the tradeable content here. A statutory line drawn at exceeding human cognitive ability [3] is hard for any agency to administer, while a one-time tax on AI equity needs only a rate and a date [12]. The counter-thesis is Sanders' own timetable, "we have to act immediately when Congress returns" [17], and the company he has been keeping while he builds it, including a conference last week alongside Republicans such as Steve Bannon [13]. A Republican co-sponsor on a ban, or Trump reversing his position on oversight [10], would undercut the case for the levy.
What to watch
- Kelly's bill text on Thursday, and whether it names a contribution rate or a base for the fund tech companies would pay into.
- Whether the bill gives the Department of Artificial Intelligence a deadline for writing safety rules, since the development pause runs until those rules exist.
- Whether more current employees at frontier labs add their names, and whether any employer responds to staff endorsing criminal penalties.