OpenAI’s opposition to California’s SB 1047 has drawn criticism from two former researchers who left the company this year over safety concerns. Daniel Kokotajlo and William Saunders say the company’s position conflicts with the public case for AI regulation made by Sam Altman, their former boss.
The dispute is not only about one California bill. It is also about where powerful AI systems should be regulated, how companies should respond when concrete rules are proposed, and whether frontier AI safety belongs primarily at the state or federal level.
Why SB 1047 became a test for OpenAI
SB 1047 is described in the source article as California’s bill to prevent AI disasters. OpenAI has decided to oppose it, and that decision has become a flashpoint because the company and its leadership have repeatedly been associated with calls for AI regulation.
Kokotajlo and Saunders argue that the moment matters because the debate has moved from broad statements about safety to a specific policy proposal. In their letter urging California governor Gavin Newsom to sign the bill, they wrote: “Sam Altman, our former boss, has repeatedly called for AI regulation. Now, when actual regulation is on the table, he opposes it.”
The two former researchers had already warned that OpenAI is in a “reckless” race for dominance. Their criticism of the company’s SB 1047 stance should be read against that background: they are not merely disagreeing with a lobbying position, but raising a broader concern about whether commercial pressure is overtaking safety commitments.
They also left room for a different outcome. In the same letter, they wrote: “With appropriate regulation, we hope OpenAI may yet live up to its mission statement of building AGI safely.” That framing keeps the focus on regulation as a way to align OpenAI’s conduct with its stated mission, rather than as a punishment for the company.
OpenAI says the issue is federal authority
OpenAI rejects the former employees’ characterization. In a statement to TechCrunch, an OpenAI spokesperson said the startup “strongly disagrees with the mischaracterization of our position on SB 1047.”
The company’s response centers on the level of government that should set the rules. The spokesperson pointed to AI bills in Congress that OpenAI has endorsed and said that “frontier AI safety regulations should be implemented at the federal level because of their implications for national security and competitiveness.”
That argument does not deny the need for AI safety regulation. Instead, it says the proper venue is Congress, not California. In OpenAI’s view, frontier AI systems raise issues that reach beyond one state, including national security and competitiveness.
This distinction is central to the disagreement. Kokotajlo and Saunders see opposition to SB 1047 as a failure to support actual regulation when it appears. OpenAI presents its stance as a disagreement about jurisdiction and policy design, while pointing to federal AI bills it has supported.
Anthropic takes a different path
The source article also contrasts OpenAI’s position with that of Anthropic. Anthropic has expressed support for SB 1047 while also identifying specific concerns and asking for amendments.
Several of those amendments have since been incorporated. On Thursday, Anthropic CEO Dario Amodei wrote to Newsom that the current bill’s “benefits likely outweigh its costs,” though he did not fully endorse the bill.
That position is more conditional than a simple endorsement. It signals that Anthropic still sees tradeoffs in the bill, but believes the version now under consideration is likely worthwhile on balance.
The comparison matters because both companies operate in the same broad frontier AI environment, yet they are taking different public approaches to SB 1047. OpenAI is opposing the bill and emphasizing federal regulation. Anthropic has supported the bill while seeking changes, and its CEO has said the benefits likely outweigh the costs.
What the disagreement reveals
The SB 1047 fight shows how quickly the AI safety debate becomes practical once lawmakers propose binding rules. High-level agreement that AI should be regulated can still leave major disputes unresolved.
Several questions sit underneath the public statements:
- Should frontier AI safety rules begin at the state level when federal rules are still being debated?
- Does support for AI bills in Congress answer criticism over opposition to a state bill?
- How should companies balance safety promises against concerns about national security and competitiveness?
- What weight should the public give to warnings from former employees who resigned over safety concerns?
The source material does not resolve those questions. It does, however, make clear that SB 1047 has become a concrete measure of how AI companies respond when regulatory pressure moves from principle to text.
For Kokotajlo and Saunders, OpenAI’s opposition is disappointing but not surprising because it fits their concern that the company is racing for dominance. For OpenAI, the criticism misstates its position because the company says frontier AI safety rules should be federal. For Anthropic, the bill appears imperfect but, after amendments, potentially worth supporting on balance.
That leaves Gavin Newsom at the center of the immediate decision described in the source article. The broader issue will remain even after this bill: AI companies are no longer being judged only by whether they say regulation is needed. They are being judged by what they do when a specific regulation reaches the table.