Some links on this page are affiliate links: if you buy through them we may earn a commission, at no extra cost to you.
Garry Tan’s May 2024 position was not “no AI regulation.” The Y Combinator president and CEO said regulation was probably necessary and supported parts of federal risk-management efforts, while calling AI bills moving through California and San Francisco “very concerning.” His objection was to rules he considered too broad, speculative or burdensome for smaller developers—not to every safeguard.
What Tan said—and when
At an Economic Club of Washington, D.C. event in May 2024, Tan said AI regulation was likely necessary. He expressed general support for the National Institute of Standards and Technology’s work on generative-AI risk management and said large parts of the Biden administration’s AI executive order were on the right track. At the same time, he criticized proposed AI legislation in California and San Francisco. TechCrunch reported his remarks on May 26, 2024.
That date matters: these comments describe Tan’s position at that event, not necessarily his view in 2026. They also need context. Tan leads Y Combinator, a startup accelerator and investor whose portfolio includes AI companies. His perspective is that of a startup ecosystem advocate, not a neutral regulator or an academic AI-safety specialist. His later Senate testimony and responses to senators’ questions show his broader emphasis on “little tech” and competition.
The Tool Desk
Outbyte Driver Updater FREEFix the driver behind crashes, sound loss and screen glitchesFind Drivers →Outbyte PC Repair FREEClear out junk files and repair common Windows errorsFree Scan →What kinds of regulation did Tan support?
NIST’s risk-management framework
Tan said he was generally supportive of NIST’s effort to develop a framework for managing generative-AI risks. The NIST AI Risk Management Framework is a risk-management resource, not a single comprehensive AI law or a universal legal mandate. Laws, contracts, procurement rules or agency policies may adopt similar practices, but the framework itself does not make every recommendation binding on every company.
#1 Best Overall
The approach discussed in contemporary coverage included applying existing privacy and copyright law, informing users when they are interacting with generative AI, and addressing harmful outputs such as child sexual-abuse material. The distinction is between building practices and standards to identify and manage risks and imposing one sweeping statutory prohibition.
The Biden administration’s executive order
Tan also said that large parts of the Biden administration’s October 30, 2023 AI executive order were probably on the right track. The order directed federal agencies toward actions involving safety and security testing, standards and guidance, privacy, civil rights, consumer protection and competition, among other subjects. It also addressed government access to information about certain advanced AI systems and support for smaller developers and researchers. The order was an executive action, not a permanent act of Congress; it could be changed or revoked by a later administration.
Which bills did he criticize?
Tan called proposed AI bills in California and San Francisco “very concerning,” but the contemporary account of his remarks does not identify a complete list of the measures he meant. It would therefore overstate the evidence to say he rejected every AI bill in either jurisdiction or opposed every provision of a specific bill.
Free tools Windows power users keep installed
One-click scans. No signup required.
Rank #2
The most prominent California example in the surrounding debate was Senate Bill 1047, introduced by state Sen. Scott Wiener. Contemporary coverage connected it to the discussion, but does not establish that Tan named the bill as his sole target or publicly opposed each of its provisions. It is safest to treat SB 1047 as an example of the kind of state-level approach he viewed skeptically.
What SB 1047 proposed—and why it was disputed
SB 1047 sought to impose safety and security responsibilities on developers of certain powerful AI models, with a focus on catastrophic-risk scenarios. Its supporters argued that developers of highly capable systems should face enforceable obligations before severe harms occur. Wiener’s case for the bill is discussed in this Lawfare interview, and the legislative debate is documented in California hearing materials.
Critics raised concerns about uncertainty, liability, enforcement, compliance costs, effects on open-source development and California’s ability to regulate frontier AI without driving activity elsewhere. Those concerns do not prove regulation would kill startups; they identify a possible distributional effect. Large companies can generally devote more resources to legal review, testing and compliance than early-stage firms. The Reuters explainer describes the competing arguments.
The dividing line in Tan’s argument
Tan’s stated preference was for a more measured approach: address concrete harms and avoid writing rules around what he characterized as science-fiction concerns that were not occurring. That is his argument about what regulation should prioritize, not proof that catastrophic risks are unimportant or impossible. The disagreement is partly about evidence and timing: whether lawmakers should impose obligations to prevent a potential severe future harm, or first focus on harms already observable.
- Existing harms: privacy violations, copyright disputes, fraud, cyberattacks, harmful sexual imagery, discrimination and unsafe deployment can be addressed through existing law or targeted safeguards.
- Frontier risks: rules for powerful future systems aim to prevent harms that may be severe but are harder to measure in advance.
- Who pays: testing, reporting and liability requirements may improve safety, but fixed compliance costs can weigh more heavily on small developers than on major labs.
Tan warned that poorly designed rules could raise barriers to entry, reduce consumer choice and strengthen the biggest AI firms by leaving fewer challengers able to comply. His stated concern was a market in which a small number of companies control models and extract rents. That concern fits YC’s interest in a viable startup ecosystem; it is relevant context, but does not by itself establish whether his policy judgment is right.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.How to assess the policy trade-off
Three questions help clarify the difference between targeted safeguards and rules that may entrench incumbents:
- What harm is the rule targeting? A rule tied to a demonstrated consumer or safety harm is easier to evaluate than one built around an uncertain scenario, although prevention can require acting before harm is widespread.
- Is it enforceable and sufficiently clear? Developers need to know which systems and actions are covered, what compliance requires, and how regulators will judge failures.
- Who bears the compliance cost? If requirements are expensive regardless of company size, they may be easier for incumbents to absorb. That does not settle whether the safety benefit justifies them, but it is part of the policy design.
Federal and state rules pose a related trade-off. State action can move ahead when Congress is slow, while a state-by-state patchwork can create conflicting obligations. California’s significance to the technology sector also means its choices may affect companies beyond the state. Neither state action nor federal action is automatically better; the question is whether the rule is clear, proportionate and capable of addressing the risk.
What changed after the 2024 debate?
In September 2024, California enacted a number of AI-related measures, while Gov. Gavin Newsom also vetoed legislation he said was not sufficiently flexible or comprehensive. His office described initiatives addressing areas including deepfakes, watermarking, children, workers and misinformation. These later actions show that the state’s AI policy did not reduce to a single frontier-model bill. They are subsequent context, not part of Tan’s May remarks. The governor’s September 29, 2024 announcement summarizes the package and initiatives.
Do these 3 things before closing this tab:
1Fix the driver behind crashes, sound loss and screen glitches2Repair Windows errors before they cause bigger problems3Scan for outdated or missing drivers - takes under a minuteQuick 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.
