US politicians shift AI worries from data centers to AI models

In short: US politics is starting to focus less on where AI runs, and more on what the AI models might do.

What's going on

Until recently, many local fights about AI were really fights about data centers, the large buildings full of computers that power AI systems. People worry about their electricity use and the strain they can put on local power grids.

Now, more attention is turning to the AI models themselves, meaning the software that produces answers, images, or other outputs (like the “brain” of the system). The Financial Times points to a new burst of public warnings from AI insiders. Jacob Coxon, described as an outgoing Anthropic employee, said colleagues believed AI “could kill us all by the end of the decade.” Anthropic’s Evan Hubinger added that he personally thought there was a greater than 10 percent chance within the next decade.

Politicians from both parties are also raising alarms. Senator Chris Murphy, a Democrat, said companies were in a “blind race to build a death machine first.” Senator Ted Cruz, a Republican, called the risks of AI models going rogue “dangerous and frightening.”

A concrete policy example is new legislation from Senator Bernie Sanders and Representative Greg Casar called the Ban Artificial Superintelligence Act. It would permanently ban “superintelligent AI” and temporarily pause advanced AI deployment until a federal regulator sets safety rules.

What to watch

The FT also describes a growing split inside the AI industry over how strict rules should be, including debates over “closed” versus “open” models (whether the system’s key settings are kept private or shared more widely). As the US midterm elections approach, watch for more candidates talking about AI risks, and for more industry money and lobbying aimed at shaping any new rules.

Source: Financial Times