Oct 12, 2026 · 57 min
Alex Bores puts catastrophic AI risk at 20%
Alex Bores Puts The Odds Of AI Doom At 20%
The conversation connects AI’s potential for catastrophic harm to concrete choices about regulation, infrastructure, public control, and electoral strategy.
- 1Alex Bores estimates a 20% chance of catastrophic AI harm if development continues on its current path.
- 2The RAISE Act illustrates how safety proposals can be weakened while lawmakers balance disclosure, audits, enforcement, and innovation.
- 3AI could become a defining political issue, reshaping U.S.-China competition and the coalitions behind the 2028 election.
Don't miss
Bores argues that a temporary pause on advanced AI development matches the scale of a 20% catastrophic-risk estimate, despite difficult enforcement and competition questions.
The brief
New York Assemblyman Alex Bores estimates a 20% risk of catastrophic harm if AI development stays on its current path, while warning that immediate disruptions also demand action.
Bores uses New York’s RAISE Act to show the gap between ambitious AI safety goals and legislation shaped by negotiation, enforcement limits, and political pressure.
The policy menu ranges from incident reporting and independent audits to data-center limits, licensing, public control, and a temporary pause on advanced development.
The conversation widens to China and U.S. competition, arguing that shared safety incentives may matter more than the assumption that an AI race is inevitable.
Bores sees AI as a potentially powerful 2028 issue because voters want elected officials to act, yet remain skeptical of both parties and technology companies.
What was said on this episode
26 statements · 15 positive · 7 negative · 2 mixed · 2 neutral
AI has approximately a 20% catastrophic-risk probability on its current trajectory.
“on our current trajectory, not counting the recent momentum and changes, like on just how we're building things out right now, probably 20.”
Listen at 4:46
Known safety interventions could reduce AI catastrophic risk below 1%.
“If there if there were actions taken that we know how to take to really change the nature of this race, less than less than one percent.”
Listen at 5:05
AI will help attackers design highly harmful biological agents.
“With biological risks, it'll help the attackers design really truly awful viruses or bacteria.”
Listen at 9:04
Biological defense advances more slowly than biological offense.
“And so defense in bio is just slower than offense.”
Listen at 9:23
AI lab safety commitments should become legally mandatory.
“I think the time for voluntary agreements is done. Like, let's just make this law and make it happen.”
Listen at 11:31
The RAISE Act was the strongest AI safety law when enacted.
“The Rays Act, when passed, was the strongest AI safety law in the country.”
Listen at 15:53
Bores continues to support the original RAISE Act draft.
“I stand by the original draft.”
Listen at 21:53
Hochul’s AI record should not alone change an AI-focused voter’s election choice.
“if AI is your issue, you know, I don't think that would be a reason to to change your position in this election.”
Listen at 22:48
Kathy Hochul’s team appears to be taking AI safety seriously.
“But they seem to be taking it really seriously.”
Listen at 23:30
AI companies should be legally required to act on significant discovered risks.
“If the AI companies discover... pretty significant risks, they need to, by law, be required to act on those”
Listen at 25:00
A one-year New York data-center moratorium was appropriate during rulemaking.
“I voted for a limited moratorium here in New York that was for one year while our electric regulator set up rules for these data centers.”
Listen at 26:12
Some opposition to data centers is specifically opposition to AI.
“And so some portion of that opposition is still specifically against AI.”
Listen at 28:12
Bernie Sanders’s advanced-AI pause proposal matches the scale of current challenges.
“I think it's meeting the scale of the challenges of our time.”
Listen at 29:07
Advanced AI development should be slowed without proof of safety.
“at some point, putting the brakes on it, if we don't have proof that we can do it safely, is the right approach.”
Listen at 30:01
More capable AI models create risky territory without sufficient safety advances.
“We are going to enter risky territory if these models continue to get more and more capable, and we have not made the sufficient advances on how to do it safely.”
Listen at 31:50
Government stock warrants could insure against highly successful AI companies.
“Having the government take options, take stock warrants on all of the in all of the companies”
Listen at 33:03
Government licensing of AI models may be appropriate, but Bores is undecided.
“Maybe. You know, I don't have strong feelings one way or the other here.”
Listen at 34:11
Voters in both parties broadly support more AI regulation and slower development.
“Large majorities of both parties, voters of both parties want there to be more regulation, think we are moving too quickly.”
Listen at 37:35
China regulates AI more extensively than Western countries.
“China regulates AI far more than anything in the Western world.”
Listen at 40:16
Stopping U.S. frontier models would substantially slow China’s AI progress.
“if our frontier models stop, that would substantially slow down China's progress”
Listen at 40:57
Parties seeking victory in 2028 should present substantive AI policies.
“whichever party wants to win in 2028 should be focused on making sure they're putting forward real AI policies.”
Listen at 43:40
Officials who publicly update mistaken AI positions are a positive signal.
“elected officials that are willing to say I made a mistake and here's what I've learned and here's how I'll change are a good sign.”
Listen at 45:32
Gavin Newsom is updating his AI views and proposing AI-safety measures.
“certainly he is making overtures to updating his beliefs based on what we've seen and trying to propose things around AI safety.”
Listen at 46:40
AI will have a disproportionate impact on upcoming midterm election results.
“AI, I think, is going to have a disproportionate impact on the midterms”
Listen at 51:36
AI’s rapid development helps prevent political positions from becoming entrenched.
“And partially, though, I think that's just because it's moving so quickly that stances don't have a chance to ossify.”
Listen at 53:52
AI could substantially disrupt existing political assumptions and alignments.
“this feels like the kind of issue that could really shake up assumptions in politics.”
Listen at 56:17
Statements are attributed to the speaker as said on the episode and reflect their view at the time, not PodLume's. They are not advice.