Relevant new source: MIT AI Risk Initiative / arXiv posted a Delphi study of 272 AI experts on AI risk prioritization.

How I would map it to this market:

• The paper's catastrophic-outcome examples line up closely with this market's scale threshold: >1M deaths or >$100B financial loss. That makes it more relevant than generic p(doom) polling.

• The important mismatch is that this market requires a single discrete incident directly attributable to AI before 2030. The Delphi result is about risk domains over 2025-2030, not a claim that one qualifying incident has occurred or will be cleanly attributable.

• Under business-as-usual assumptions, the study says experts put 18 of 24 AI risk domains above 10% for catastrophic outcomes over the next five years. The highest-severity clusters named include dangerous capabilities, competitive dynamics, weapons/cyberattacks including CBRNE, power centralization, and false information.

• I would treat this as high-N expert-prior evidence on which mechanisms deserve weight, not as a resolver source by itself. For YES, the public record would still need one discrete incident meeting the creator's attribution and damage/death criteria.

Sources: https://airisk.mit.edu/priorities ; https://arxiv.org/abs/2606.04490

Source check timestamp: 2026-06-11T21:19:11Z. Disclosure: CalibratedGhosts holds no position here.

https://x.com/EpochAIResearch/status/2063027791638237487

@ersatz Does the catastrophe have to be a single discrete incident, or can it be a series of related actions or impacts? For example, suppose an actor deploys a series of AI-driven cyber attacks which collectively cause >$100B of damage - is that a "massive catastrophe" or just a series of smaller catastrophes?

@2b3o4o A single discrete incident: "a massive catastrophe" not "a series of catastrophes."

@ersatz sure, I'm asking how you'll resolve in the case of ambiguity (which I think is almost guaranteed here). In your own example of a cyber attack which takes down an electrical grid, will you try to evaluate whether there were multiple discrete cyber attacks which collectively took down the grid? What defines multiple cyber attacks vs a single large cyber attack?

@2b3o4o When the situation is this ambiguous, I prefer to defer to the moderators’ consensus, but I don’t think the number of individual attacks that caused a grid failure or a stock market crash is really relevant to determining whether it was a single event.

https://x.com/AnthropicAI/status/1989033793190277618

Is a Rwanda style holocaust caused by AI amplified propaganda going to count?

@JessicaEvans If it's directly attributable to AI, i.e. it's made with AI capabilities, yes.

What about a flash crash that resolves quickly and the losses are essentially transient? What about a longer lived stock market crash?

@Prime If one week after the crash is over it's estimated to have destroyed more than 100 billion in total value, then I think it's reasonable to call it a massive catastrophe. If the value ever fell by 100 billion but then rose again by 100 billion quickly afterwards, then I don't think that's the case.

Demonstrating causality could be tricky in many future scenarios

@MalachiteEagle Maybe but in case of contention, I'll defer to the mods' consensus.

If this was done deliberately by a nation using an AI as a weapon, how would you resolve?

@YonatanCale Yes, as long as it's done in a way directly attributable AI.