Quick wins for a faster PC:
Repair Windows errors before they cause bigger problemsFix Now →Scan for outdated or missing drivers - takes under a minuteDriver Scan →Clear out junk files and repair common Windows errorsFree Scan →Former OpenAI employees William Saunders and Daniel Kokotajlo criticized the company in August 2024 for opposing California’s frontier-AI safety bill, SB 1047. In a letter shared with Politico, they argued that OpenAI had publicly supported AI regulation but resisted it when lawmakers proposed concrete requirements.
OpenAI said that characterization was inaccurate: its objection was to California regulating frontier AI at the state level, not to AI regulation generally. The dispute became historical on September 29, 2024, when Governor Gavin Newsom vetoed SB 1047 after it passed both chambers of the Legislature.
| # | Preview | Product | Price | |
|---|---|---|---|---|
| 1 |
|
California School Law: Third Edition | $43.16 | Buy on Amazon |
| 2 |
|
California Landlord's Law Book, The: Rights & Responsibilities | $34.88 | Buy on Amazon |
| 3 |
|
Merriam-Webster's Dictionary of Law, Newest Edition, Trade Paperback | $10.99 | Buy on Amazon |
| 4 |
|
California Statutes - Evidence Code 2026 | $19.99 | Buy on Amazon |
As an Amazon Associate I earn from qualifying purchases.
What the former employees said
Saunders, formerly a member of OpenAI’s technical staff, and Kokotajlo, formerly a member of its policy staff, left the company in 2024 after raising concerns about its approach to AI safety. Their letter criticized OpenAI’s opposition to the Safe and Secure Innovation for Frontier Artificial Intelligence Models Act, commonly called SB 1047.
Their central argument was an apparent gap between public statements and corporate action. They said Sam Altman had previously called for government involvement in AI regulation, while OpenAI opposed what they described as relatively limited safety obligations once a specific bill was before California lawmakers.
#1 Best Overall
They also argued that several proposed requirements resembled safety practices AI companies, including OpenAI, had already said they would undertake voluntarily. Their concern was that voluntary commitments could later be weakened, abandoned or enforced inconsistently.
Those claims came from former employees who were critics of the company, not from neutral outside auditors. They are important allegations about OpenAI’s priorities, but they should be understood as their interpretation of events.
What SB 1047 would have done
SB 1047 was not a law covering every AI system or consumer chatbot. The enrolled version focused on developers of the most computationally expensive and capable AI models, along with some companies supplying the computing resources used to train them. The final bill text centered on preventing specified catastrophic harms from frontier systems.
The Tool Desk
Outbyte PC Repair FREEClear out junk files and repair common Windows errorsFree Scan →Outbyte Driver Updater FREEScan for outdated or missing drivers - takes under a minuteDriver Scan →In practical terms, the bill would have required covered developers to:
- conduct pre-deployment safety and risk assessments;
- maintain written safety and security protocols;
- take reasonable care to prevent or reduce specified catastrophic risks;
- maintain the ability to halt or disable a covered model or training process when necessary;
- submit to third-party auditing and government oversight; and
- provide protections for employees reporting safety concerns.
The bill also created potential liability for covered developers and certain participants in the AI-development chain. Its obligations could therefore have affected not only model creators but, depending on the provision, companies modifying, deploying or providing computing resources for covered systems.
The phrase “kill-switch bill” became common shorthand, but it is imprecise. SB 1047 did not require a universal consumer-facing button on every AI product. It addressed a developer’s capability to stop or disable a covered model or training operation under specified circumstances.
OpenAI’s stated objection
OpenAI said Saunders and Kokotajlo mischaracterized its position. According to the company’s position reported by TechCrunch, frontier-AI rules should be enacted at the federal level because they involve national security and U.S. competitiveness.
That distinction matters. The accurate description is:
- Too broad: OpenAI opposed AI regulation.
- Accurate: OpenAI opposed SB 1047.
- More precise: OpenAI preferred federal regulation of frontier AI over California’s state-level approach.
OpenAI also pointed to AI legislation in Congress that it had supported. Its position reflected concerns that a single-state regime could create fragmented obligations for companies operating nationally and internationally.
The policy questions behind the dispute
The argument was not simply “AI companies versus safety advocates.” It involved at least four separate questions:
- Is regulation needed? Former employees and OpenAI could agree that advanced AI creates serious risks while disagreeing about the appropriate rules.
- Who should regulate? OpenAI favored federal action, while supporters of SB 1047 argued that California could act while federal legislation remained incomplete.
- Is model size a useful proxy for risk? The bill emphasized computing and model capability. Critics argued that dangerous outcomes can depend more on deployment, access and use than on training scale alone.
- Would liability improve safety? Supporters viewed enforceable duties as a counterweight to commercial pressure. Critics warned that uncertain liability could discourage research, open-source development or deployment.
Supporters of the former employees’ position argued that companies have incentives to release increasingly capable systems quickly and that voluntary commitments may not survive changing commercial priorities. OpenAI and other critics argued that compliance fragmentation, unclear liability and state-level regulation could create substantial costs without providing a reliable safety framework.
Do these 3 things before closing this tab:
1Repair Windows errors before they cause bigger problems2Scan for outdated or missing drivers - takes under a minute3Clear out junk files and repair common Windows errorsWhy Anthropic took a different position
Anthropic did not simply endorse the original bill without reservations. It raised concerns and requested amendments, including questions about liability, open-source or open-weight models, covered-model thresholds and the discretion given to regulators.
After changes were made, CEO Dario Amodei said the bill’s benefits likely outweighed its costs, while acknowledging uncertainty and unresolved ambiguities. Anthropic’s position therefore illustrated a qualified judgment about the amended legislation, not blanket approval of every version of SB 1047.
The split showed why the industry debate could not be reduced to whether a company supported “AI safety.” Companies differed over jurisdiction, implementation, liability and how to identify genuinely dangerous systems.
What happened to SB 1047
California lawmakers passed the bill after the controversy reported in August 2024. The Assembly approved it 48–16 on August 28, and the Senate approved it 30–9 on August 29. It was presented to Governor Newsom on September 9, according to the official bill history.
Newsom vetoed the bill on September 29, 2024. SB 1047 therefore did not take effect and is not current California law.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.Why Newsom vetoed the bill
Newsom’s veto message did not say that AI safety regulation was unnecessary. Instead, he argued that SB 1047 focused too heavily on the size of an AI model and did not sufficiently account for where and how a system was deployed.
He specifically pointed to questions such as whether a system was:
- deployed in a high-risk environment;
- used for critical decision-making; or
- handling sensitive data.
In his view, a smaller model used in a sensitive or consequential setting could create serious risks, while a very large model might not present the same danger in every deployment. His criticism was therefore directed at the bill’s risk framework, not necessarily at the broader goal of imposing safeguards.
Newsom also announced other California AI initiatives and signed separate measures concerning areas including deepfakes, watermarking, performers’ digital likenesses and election-related synthetic media. Those actions showed that the veto was not a declaration that California should take no action on AI; they reflected a different approach to regulating specific harms and uses.
What the controversy revealed
The former employees’ criticism exposed a persistent tension in AI policy: voluntary safety commitments can be faster and more flexible, but they may be difficult for the public to verify and enforce. Legislation can create accountability, but poorly designed rules may impose uncertainty, miss important risks or regulate a proxy rather than the harmful use itself.
SB 1047 failed as legislation, but its underlying questions remain. Should frontier-AI safeguards be set nationally or by individual states? Should rules focus on the model, the developer, the computing infrastructure or the deployment context? And how can regulators create enforceable duties without making responsible research and development impractical?
The most accurate account of the 2024 dispute is therefore narrower than either side’s strongest slogan. Saunders and Kokotajlo accused OpenAI of retreating from its earlier support for regulation. OpenAI said it supported regulation but opposed California’s state-level approach. Anthropic supported the amended bill in principle while retaining reservations. Newsom ultimately vetoed it because he believed its model-size framework did not adequately capture real-world risk.
Free tools Windows power users keep installed
One-click scans. No signup required.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
