Meet the Politician the AI Industry Is Trying to Stop

Episode Summary

Executive Summary: Odd Lots interviews New York Assemblymember and congressional candidate Alex Bores about AI regulation, his anti-AI-super-PAC targeting, and his Palantir/government-tech background. The conversation centers on his RAISE Act, which would impose safety, disclosure, and testing requirements on frontier AI labs, while also touching on deepfakes, scams, education, energy, crypto regulation, and how government can use data better.

Main Topics: AI regulation and the RAISE Act (Priority: 5/5): Bores explains his bill to require major frontier AI labs to publish safety plans, disclose serious incidents, and stop releasing models that fail safety tests. He frames it as a practical floor for safety rather than a blanket ban. Political backlash and AI industry spending (Priority: 5/5): The hosts and Bores discuss a new pro-AI super PAC that is spending millions to defeat him because of his regulatory stance, highlighting how AI has become a major political battleground. OpenAI, Palantir, and the tech-policy nexus (Priority: 4/5): Bores discusses his past at Palantir and how that experience shaped his focus on implementation, data integration, and government effectiveness, while the hosts note the irony of a tech-savvy regulator being targeted by tech interests. Deepfakes, scams, and low-level digital harm (Priority: 4/5): The episode broadens from frontier-model safety to everyday harms like AI-generated porn, scam texts, fake books, and deepfake images/audio, arguing that ordinary users are already feeling AI’s negative externalities. State vs federal authority over AI (Priority: 4/5): Bores criticizes Trump’s executive order preempting state AI regulation and argues that states like New York are leading on chatbot disclosures, self-harm alerts, and deepfake protections. AI’s broader public-policy footprint (Priority: 3/5): The discussion extends to labor, education, energy demand, water use, inequality, and cybersecurity, with Bores arguing AI could be transformative but needs democratic oversight and standards. Using data to improve government performance (Priority: 3/5): Bores describes how he measures whether legislation works after passage, using examples like telemarketing fines and moped registration, and argues government should be more iterative and evidence-based.

Key Arguments: AI should be regulated with targeted safety standards, not left entirely to market incentives; otherwise companies may rush unsafe systems to market. The RAISE Act is designed to affect only a small number of frontier labs, using compute thresholds and safety incidents as triggers. Public safety plans, incident disclosure, and mandatory testing are analogous to past regulatory responses to dangerous products like tobacco. State regulation is necessary because federal preemption may block useful protections on chatbots, self-harm warnings, and deepfake harms before Congress acts. Deepfake and scam harms are not hypothetical; they are already eroding trust and causing real-world abuse, so enforcement and provenance standards matter now. Government should evaluate laws by outcomes, not just passage; Bores repeatedly emphasizes implementation, data tracking, and iterative correction. AI could deliver major benefits in medicine, education, and productivity, but only if the public retains a voice in how it develops. The industry’s objection that regulation will crush innovation is overstated; Bores argues compliance burdens are small for major labs and were estimated by lobbyists as minimal.

Data Points: Bores's legislative output: 27 bills - He says he passed 27 bills in three years in the New York State Assembly. Congressional comparison: 27 bills - He notes this equals the number passed by Congress as a whole in 2023. RAISE Act compute threshold: $100 million - Companies that spent at least $100 million specifically on compute for a final training run would be covered. Frontier-model complexity threshold: 10^26 flops - One definition of a frontier model under the bill is training with 10 to the 26 FLOPs. Distillation threshold: $5 million - A model trained via knowledge distillation with at least $5 million spent would also be covered. Fines under the bill: $10 million first violation; $30 million subsequent violations - Bores cites these penalty levels, while saying he personally thinks they are too low. Original fine concept: 10% of training costs - He says an earlier version scaled penalties to a company’s training costs. Lobbyist estimate of compliance: 1 extra full-time employee - Bores says labs’ own estimate was that Google or Meta would need one additional FTE to comply. Super PAC spending on Bores: $10 million - He says the PAC first announced plans to spend multiple millions, then increased the target to $10 million. PAC total planned spend: $100 million - Bores jokes he hopes to use up the PAC’s reported $100 million plan. Chatbot disclosure rule: Every 3 hours - New York regulations would require chatbots to identify themselves at the start and every three hours of continuous conversation. Telemarketer fine effect: 4x as many fines - Bores says raising the statutory maximum fine led to four times as many fines being issued. Moped registrations at bill start: 1,700 - He cites the number of mopeds registered when the point-of-sale registration bill took effect. Moped registrations one month later: 1,400 - The number registered later fell, suggesting re-registration, not initial registration, was the issue. AI PAC donors: $5.5 million / $1 million / at least $2.5 million - Hosts and Bores cite Mark Andreessen, Joe Lonsdale, and Greg Brockman donations tied to pro-AI political efforts.

Pivotal Quotes: "The RAISE Act is squarely within that realm." — Alex Bores: He is describing the bill as a balanced approach between innovation and safety. "It is the technology that has the widest bounds of what could potentially come from it." — Alex Bores: He explains why AI needs proactive policy rather than laissez-faire development. "The only way to solve that problem is you need actual enforcement. You need there to be consequences." — Alex Bores: He is discussing scams, low-trust behavior, and the need for meaningful penalties.

Implications: The episode frames AI as a central 2026 political issue. For listeners, it shows that regulation is moving from abstract debate to concrete state-level rules, with real fights over safety, industry power, and who gets to set the standards.