Major publicly traded cybersecurity firms lost roughly $65–80 billion, or about 8–10% of their combined value, in the days following disclosure of the Hugging Face/OpenAI incident; by early September they had recovered roughly $58 billion, representing about 70–90% of that drawdown, depending on whether July 15 or July 20 is used as the pre-event baseline.

That figure comes from GPT Pro, with more detail at the link. As a very rough approximation, suppose you dismiss the price bounceback altogether as either random or attributable to good earnings reports. You are then left with "the value of previous cybersecurity efforts" falling by eight to ten percent. That is very broadly consistent with some of the estimates discussed in my previous post on the numbers. This is essentially the logic of an event study, a standard tool in financial economics: measure how a category of stocks moves around a specific news event, relative to the broader market, to infer how investors priced the news. The approach has well-known limitations — other news, earnings reports, and macro moves can contaminate the window, which is precisely why the choice of baseline dates such as July 15 versus July 20 matters here.

In any case, it is a significant sum. But note that if AI models were truly on the verge of doing terrible things to us, the market might have estimated the value of our cyberprotection as falling by more than eight to ten percent. Conversely, the fact that much of the drawdown was recovered within weeks suggests investors did not treat the incident as a durable revaluation of the cybersecurity sector, whether because they judged the risk contained or because subsequent earnings dominated the picture.

More generally, perhaps these numbers could be used to discipline the discussion a bit? Or will I instead read long lists of reasons why they show us nothing — in which case, try coming up with some other market price-based indicators of AI risk? The VIX will not do it for you, not these days. Other candidates exist, such as prediction-market odds on AI milestones, the valuations and share prices of major AI developers and chipmakers, or the pricing of cyberinsurance — each with its own complications, but each at least producing a number. I see many metaphors, insinuations, and random anecdotes of AI terror, but not numbers. Maybe you think your ideas about AI risk are so important that no market prices can reflect them? (If you really believe that, does it mean you would not be worried, and would not cite the numbers, if the value of those companies fell by ninety percent?)

I am sure others can improve on what I am putting forward, and we should also track the continuing progress of these share values over time, especially if other AI hack attacks surface.

Overall, I am extremely skeptical of arguments that essentially take the form of "what I am concerned about is too big and too important to show up in any market prices." Pick your market prices!

Source: Marginal Revolution