AI news for Saturday, November 7

OpenAI and Anthropic each put up a paper on October 26 claiming an internal AI model had disproved the Exponential Time Hypothesis, a 1999 guess about how hard 3-SAT is. Both come with Lean proofs, and no human has confirmed the result yet. Impagliazzo and Paturi, who proposed the hypothesis, are among the people leading the check.

On November 5, federal prosecutors in Manhattan unsealed an indictment saying three men took about $129 million from companies by faking their chief executives' voices with AI. The callers posed as the boss, then handed staff off to fake lawyers at real law firms, who asked for the wire. The men are presumed innocent.

Mustafa Suleyman, who runs Microsoft AI, called a Delaware plan to let AI agents run companies deeply concerning. The proposal would create a new kind of company managed by an agent, inside a 30-month sandbox, with its human owner shielded from its debts. James Comer, who chairs the House Oversight Committee, is already investigating it.

Sam Altman posted on X on November 3 that people should always be in control of the future, and that OpenAI is against the idea of people doing whatever AI decides is worth rewarding. The phrase he put in quotation marks comes from Machines of Loving Grace, the essay Dario Amodei published in October 2024, and Altman didn't name him.