California state Senator Scott Wiener is calling for stronger federal oversight of artificial intelligence after winning the November 3, 2026, congressional election in San Francisco. The Democrat, who will succeed former House Speaker Nancy Pelosi, says recent developments in advanced AI have increased the urgency of establishing safeguards before the technology becomes more difficult to control.
Wiener expressed particular concern about new research suggesting that artificial intelligence is helping developers create more capable systems at an accelerating pace. He believes Congress should establish enforceable safety standards, protect employees who report dangerous practices and preserve the ability of states to regulate AI independently.
His election places a prominent advocate of AI regulation in the House of Representatives as lawmakers face competing demands from technology companies, researchers and communities affected by the rapid expansion of AI infrastructure.
New Research Raises Questions About AI Improving Itself
Wiener's concerns intensified following a November 5 report from METR, an independent nonprofit organization that evaluates artificial intelligence capabilities and risks. Researchers examined how Anthropic, the developer of the Claude AI models, uses artificial intelligence to support its own research and development.
The study found that Anthropic's models appeared to be improving on selected benchmarks at approximately 1.9 times their previous rate. Researchers estimated that AI-assisted development was responsible for a substantial portion of that acceleration.
The findings followed an assessment conducted between August 12 and October 29. METR researchers received access to internal information about Anthropic's development process, including model evaluations, training methods and interviews with company employees.
The results have attracted attention because they suggest AI systems are becoming increasingly useful in developing more advanced versions of themselves.
This process is sometimes described as recursive self-improvement. In its most advanced form, it could involve AI systems independently designing, testing and improving successor models with limited human participation.
However, the METR findings do not establish that Anthropic has achieved fully autonomous recursive self-improvement. Researchers said the company's current progress still depends heavily on human researchers using AI tools, rather than AI systems independently carrying out the most important stages of development.
METR also emphasized considerable uncertainty in its estimates. Different methods of analyzing the available evidence produced different acceleration rates, while some important internal findings could not be publicly disclosed because Anthropic requested redactions.
The organization warned that continued progress could create additional risks if increasingly capable systems begin accelerating research faster than safety evaluations can keep pace.
Wiener Wants Congress to Act Before AI Risks Increase
Wiener discussed the METR findings in an interview with MIT Technology Review shortly after his congressional victory. He described the report as concerning and argued that lawmakers were running out of time to establish meaningful oversight.
His proposals include stronger whistleblower protections, independent safety testing and auditing requirements for companies developing advanced AI systems. He also supports mechanisms that would allow potentially dangerous systems to be shut down in an emergency.
Such mechanisms are often called kill switches, although their practical design would depend on how individual AI systems operate. Creating a reliable shutdown process for highly autonomous technology could present technical and regulatory challenges.
Wiener has also advocated creating a dedicated federal regulatory body to oversee advanced AI safety and security. Such an agency could establish standards, investigate serious incidents and monitor compliance with federal requirements.
Supporters of stronger oversight argue that companies developing increasingly capable systems should face independent scrutiny rather than relying exclusively on voluntary internal testing.
Critics of expansive regulation raise different concerns. Some technology companies and industry groups argue that complicated approval processes and overlapping rules could slow innovation, increase compliance costs and make it harder for smaller developers to compete.
Congress would need to address those competing considerations when deciding whether to establish a national regulatory system. Wiener has indicated that he wants requirements with meaningful enforcement powers rather than measures limited to voluntary commitments.
California's SB 53 Provides a Legislative Starting Point
Wiener enters Congress with experience negotiating legislation covering advanced AI systems. His first major effort, Senate Bill 1047, passed the California Legislature in 2024 but was vetoed by Governor Gavin Newsom.
The proposal would have imposed safety obligations on developers of particularly powerful AI models. It also sought to establish protections against severe harm involving advanced systems.
Newsom rejected the legislation in September 2024, arguing that its approach focused too heavily on the size and cost of AI models rather than the actual risks posed by specific applications.
Wiener subsequently introduced a narrower proposal, Senate Bill 53, which Newsom signed on September 29, 2025. The measure became California's Transparency in Frontier Artificial Intelligence Act.
The law requires certain large AI developers to publicly disclose information about their safety and security frameworks. It also establishes procedures for reporting critical safety incidents and protects employees who disclose significant health or safety concerns.
Unlike the earlier proposal, SB 53 places greater emphasis on transparency and accountability rather than broad requirements for developers to prevent specified catastrophic outcomes.
The legislation also established the CalCompute initiative, intended to support public access to computing resources for researchers and smaller organizations.
California's experience provides a possible model for federal action, although Congress could choose different requirements. Wiener has expressed interest in expanding the principles behind SB 53 while strengthening oversight as technology advances.
Campaign Funding Raises Questions About Industry Influence
Wiener's election campaign also became part of a larger debate over the role of technology companies in shaping AI legislation.
Public First Action, a political organization backed by Anthropic, spent approximately $3 million supporting his congressional campaign. The group promotes candidates who favor stronger safeguards for advanced artificial intelligence.
That support attracted criticism from his opponent, San Francisco Supervisor Connie Chan, who questioned whether politicians receiving backing from AI-related interests could regulate the industry independently.
Wiener has defended his record, arguing that he began pursuing AI regulation before the current political spending campaigns developed. He has pointed to his legislative work in California as evidence of his commitment to establishing safety standards.
The funding also illustrates differences within the technology industry. While some companies and investors oppose extensive government oversight, others support particular forms of regulation that they argue could improve public trust and reduce serious risks.
Those disagreements complicate efforts to describe the industry as holding a single position on AI policy.
Chan conceded the congressional race following the November 3 election. Wiener will represent California's 11th Congressional District, which covers most of San Francisco, when the new congressional term begins in January 2027.
Federal Rules Could Conflict With State AI Laws
One of the major issues facing Congress is whether federal AI legislation should limit the authority of individual states.
Wiener opposes broad federal preemption that would prevent states from adopting their own safeguards. He believes state governments should retain flexibility to address emerging problems when Congress has not established adequate protections.
Supporters of nationwide standards argue that a consistent regulatory framework could help companies avoid conflicting requirements across different jurisdictions. Businesses operating in multiple states may face additional costs when laws impose different reporting, testing or disclosure obligations.
Opponents of sweeping preemption argue that federal rules could become too weak or slow to respond to new risks. They also point to states such as California, where lawmakers have already enacted measures covering advanced AI development.
The disagreement reflects a broader question about how technology regulation should be divided between Washington and state governments.
Wiener has criticized President Donald Trump's approach to catastrophic AI risks and expressed concern about the possibility of federal policies weakening existing state protections.
He has also called for greater international cooperation, including discussions involving China, because advanced AI development extends across national borders.
However, agreement on international oversight could be difficult given disagreements about national security, technology competition and access to advanced computing resources.
Data Center Opposition Adds Another Dimension
The political debate about artificial intelligence extends beyond advanced models and the possibility of catastrophic technological failures.
Across the United States, communities are questioning the construction of large data centers needed to operate AI systems. Residents have raised concerns about electricity consumption, water availability, environmental effects and potential increases in utility costs.
Virginia has become a major focus of those concerns because of its large concentration of data centers. Projects in the region have generated debate over transmission infrastructure, backup generators and who should pay for electricity network upgrades.
Some communities have pursued restrictions or temporary development pauses while reviewing the effects of additional facilities.
Wiener has acknowledged that concerns about local infrastructure can differ substantially from his focus on advanced AI safety. Residents opposing a data center may be more worried about noise, land use and power bills than the possibility of AI systems becoming difficult to control.
Those concerns are not necessarily contradictory. They represent different consequences of AI development, ranging from immediate effects on communities to uncertain risks associated with future technological capabilities.
Federal lawmakers will face pressure to address both sets of issues, although energy regulation, environmental permitting and advanced AI oversight involve different agencies and legal frameworks.
Congress Faces Difficult Choices on AI Oversight
Wiener's transition from the California Senate to Congress comes as federal lawmakers consider how to respond to rapid advances in artificial intelligence.
His priorities include mandatory safety evaluations, whistleblower protections, incident reporting and mechanisms for stopping dangerous systems. He also wants states to retain regulatory authority while Congress develops national safeguards.
The recent METR assessment provides evidence that AI-assisted research is accelerating progress at Anthropic, but it does not establish that uncontrolled autonomous improvement has begun. That distinction is important because proposed regulations must address demonstrated capabilities while accounting for uncertain future risks.
Wiener will enter the House in January 2027, where any major legislation would require support from other lawmakers and passage through the federal legislative process.
The immediate question is whether Congress can agree on enforceable requirements for advanced AI developers while addressing concerns about innovation, federal authority and the growing physical infrastructure required to support the technology.