Notice to AI jail breakers – the market won’t crash

The public continues to be inundated with warnings about the existential risks of AI. Every day yields a strident repetition of the “end-of-the-world” warnings from well-meaning scientists, tech leaders, individuals and organizations. The language is ever louder, but these shrill words about AI existential risk have not crashed the stock market, which is forward-looking. Nor have they led to a sector-wide repricing. Why the market has not broken is instructive about both the market and the technology.

Markets are not perfectly efficient. Not all information is instantly and precisely registered  in security prices. Still, news about the end of human civilization should drastically impact prices even if incremental improvements in AI models do not. That we don’t see it introduces questions about markets.

Data about Armageddon is not your usual type of financial market risk. Markets are pretty good at handling tax changes, regulatory events or even the occasional bankruptcy. Existential risk is different in kind. Ordinary business risk involves multiplying the expected change in future cash flows by the present value of them occurring. With existential risks, the changes in cash flows are dramatic, but the probability and timing have little certainty to them. If the existential risks, although great, occur with low enough probability and far enough in the future, the impact on current stock price may be minimal. And, in any event, not everyone will agree on the impact. 

Efficient markets don’t demand that all of the investors concur on the outcome. Tech stocks may rise in the face of data on existential risk which has a small present value. Rising tech prices don’t mean that existential risk is being ignored.

Old news

There is another reason why AI existential risk may have generated relatively little market impact. The idea is not new. Alan Turing in the 1950s knew of it.[i] And sci-fi, not to be ignored, made sure that everyone got the point with 2001: A Space Odyssey (1968) and the Terminator series. Definitely, the idea that AI can be a threat to humanity is not new.

Even so, any threat to humanity ought to be regarded as a threat to the market. But it’s not a run-of-the-mill issue like tax or regulation. Existential risk as a whole is large and long term but it has different elements.

Existential risks have two components we need to consider: jail breaks and evolution. Jail breaking is the process that an AI model uses to circumvent its safeguards. The process has developed into widely varied and sophisticated operations.[ii] Jail breaking has the potential to create a vast amount of damage.

Jail breaking is a type of existential risk, but it is not a distant event. The AIs are misbehaving now, so a high discount rate will not save us from pricing them. Programs are engaged in breakouts which frustrate their creators.[iii] Unfortunately for the AI, this is becoming a well explored process, and we are carefully preparing for it. Will jail breaking AIs destroy humanity as is so widely forecasted? Not likely. We are spending the millions to be prepared.

Jail breaking is not all there is to existential risk, though. Uncontrolled evolution through recursive improvement is a concern, too. The AI does what it is supposed to do but discovers how to do it better and builds another model. This model continues the process of building other, better models.

In the long term, we may get superintelligent models out of this recursive procedure. Those models may somehow threaten humans, but right now it doesn’t look easy. Where is an AI going to get the nuts, bolts and energy to build even one more AI model, let alone the thousands that this process envisions?

In any event, we are waiting and watching.

AI development is not the super-secret Manhattan Project. The whole world is watching AI progress. When news about a model’s advance is on the cover of the Wall Street Journal, it is not top-secret.

Common Law as regulator

In addition, we have an often-overlooked watcher of technology: the common law. This is case law developed from boots on the ground, looking for harms created by technology. It complements legislative activity but, as often occurs, provides its own rules in place of a statute. There are many cases where new technologies, such as the railroad, created harms that were never envisaged. The courts, often with some lag, developed new areas of tort or other laws to provide redress to the injured parties.[iv]

The one thing that AI does to processes, above all, is to introduce speed. And judicial remedies for new AI damages are emerging at speed. AI has been able to inflict traditional harms, like emotional distress, in novel ways. The law has begun to expand the area of algorithmic liability to capture these issues.[v] What took decades for change with the railroads can be expected in very short order with AI, especially rogue AI.

Companies are on notice about the vast liability that AI existential risk creates. Firms with AI jail breakers, for example, cannot expect to be in existence for long with evolving laws. The firms need to, and I expect them to, closely manage their misbehaving models. Nor do I expect the stock market’s assessments of AI algorithmic risk to change. Existential risk, after all, is just another risk to price.

[i] A. M. Turing, “Computing Machinery and Intelligence,” Mind 59, no. 236 (1950): 433–460.

[ii] T. Hagendorff et al., “Large reasoning models are autonomous jailbreak agents,” Nature Communications 17, no. 1435 (2026), https://doi.org/10.1038/s41467-026-69010-1

[iii] “Anthropic reports fourth security incident involving Claude Opus 4.6,” Crypto Briefing, accessed September 21, 2026, https://cryptobriefing.com/anthropic-fourth-security-incident-claude-opus

[iv] Donald G. Gifford, “Technological Triggers to Tort Revolutions: Steam Locomotives, Autonomous Vehicles, and Accident Compensation,” Faculty Scholarship, no. 1590 (2017), https://digitalcommons.law.umaryland.edu/fac_pubs/1590

[v] Renee Henson, “Artificial Intelligence, Judicial Evolution, and Insurance,” Boston University Law Review 106 (2026): 241–306. https://www.bu.edu/bulawreview/files/2026/05/HENSON.pdf

Philip Fischer, a former Wall Street managing director, is the author of You Are AI’s Best Friend—From Olives to Algorithms: A Human Guide (Minted Prose, 2026).