The AI existential-risk debate has moved from labs into US politics

Researcher Jacob Coxon's resignation from Anthropic on September 8 triggered a visible wave of warnings about losing control of advanced AI. Anthropic CEO Dario Amodei then called for slower development at the frontier, while more than 1,000 employees at leading AI companies signed a petition for deliberate pacing.

The safety dispute acquired names, signatures, and political attention

Coxon, who had also worked at OpenAI, left the industry arguing that leading labs were moving toward self-improving superintelligence faster than they could manage the risks. Amodei subsequently made a public case for slowing progress at the most capable frontier.

In his commentary, Zvi Mowshowitz maps how politicians from both US parties began addressing the issue. He also describes Donald Trump's sharp reversal on September 19, when Trump labeled concerns around AI another political hoax and linked them with opposition to data centers.

A technical question is becoming a fight over power and coordination

Labs can introduce evals, access restrictions, and outside scrutiny on their own. Coordinated pacing, however, runs into commercial competition, the race with China, and concerns about antitrust rules. Every actor may have a reason to slow down while fearing that a rival will continue.

For companies, this is a governance problem with immediate consequences. Voluntary commitments can alter model releases before legislation exists, but without clear rules they remain reversible and dependent on the leadership of a few labs.

Loud agreement has not produced a shared plan

Agreement that the risk deserves attention does not settle the response. Options include pre-deployment evals, outside evaluators, corporate liability, international agreements, and unilateral limits on development. Each carries a different legal and political effect.

The source is an opinionated roundup that openly takes a strong position on existential risk. It is useful as a map of actors and arguments, not as a neutral estimate of catastrophe probability.

Rules for the next models will reveal whether the shift is real

Watch whether the United States requires pre-deployment assessment, protects safety coordination from antitrust disputes, and embeds independent evaluators in development. The labs offer a second signal: whether calls for pacing alter release schedules, access rules, and incident disclosure.

Lilith's verdict

Once model pacing is decided on an election stage, the labs no longer hold the whole steering wheel. Politicians will now either build guardrails or paint another lane for the race with China.