A group of senior executives is examining a strategic acquisition proposal in a conference room at the headquarters of a large technology corporation. A huge language model created the PowerPoint deck after analyzing 40,000 pages of documents, synthesizing rival positioning, running fourteen risk scenarios, and generating a recommendation with a confidence interval.
A real-time, AI-generated earnings prediction is displayed on the CFO’s screen. An hour will pass throughout the conversation. It took almost four minutes to complete the algorithms. In situations like that, when human judgment ends and computer input begins, nobody is really certain. The market, observing from the outside, is also becoming less certain.
Algorithmic governance is not something that will happen in the future. The use of AI tools in pricing, risk assessment, capital allocation, and strategy planning at businesses big enough to have the necessary infrastructure is already occurring, albeit gradually. What this does to stock behavior at scale is less well known.
The CEO premium, which is based on the notion that certain human leaders add or detract value that isn’t shown on the balance sheet through their relationships, reputations, and judgment, has always been factored into the standard equity market model. Steve Jobs’s return to Apple was welcomed by the market not because his presence immediately altered the company’s finances but rather because it altered the notion of what Apple could achieve. It is challenging for an algorithm to produce that kind of intangible premium, and it is also challenging for another algorithm on the trading side to value.
In theory, it is replaced with a sort of simplified fundamental price. The market should be more concerned with the accuracy of the inputs and the well-defined outputs if an algorithm is executing the capital allocation plan than with the identity of the person seated in the corner office. Numerous academics have discovered evidence that algorithmically regulated corporations exhibit less “stock price synchronicity”—that is, their prices fluctuate more in reaction to company-specific fundamentals than to sentiment fluctuations across the industry.
That’s the hopeful explanation. This precision cuts both ways, according to the negative interpretation. The stocks fell in after-hours trading even though the results would have satisfied any human analyst using judgment about context and trend. Broadcom and CrowdStrike both produced strong growth numbers in recent quarters, but their guidance fell slightly short of what automated analyst models predicted. There is no curve in the algorithm’s grading system.
It’s important to give the flash crash issue more consideration than most corporate governance talks do. The 2010 Flash Crash, which was thoroughly detailed in the joint SEC/CFTC report, demonstrated what happens when automated trading systems interact with one another in ways that were not anticipated by their individual designers. A series of automated sell decisions caused equity values to drop by almost a trillion dollars for a brief period of time before they recovered in a matter of minutes.
Trading algorithms were implicated in that event. An AI governance system that misinterprets market signals, executes a significant strategic move based on flawed input, and sends automated trading systems into a self-reinforcing loop that the humans ostensibly in charge are unable to stop quickly enough is the scenario that keeps some risk analysts up at night. However, it is caused by algorithmic executive decisions.
The regulatory framework’s position on all of this is still up in the air. The SEC has been looking into transparency rules for AI’s role in significant business decisions, which is a sensible starting step. However, the issue of accountability—who is responsible when an algorithm makes a choice that reduces shareholder value—remains really unanswered.
The market’s conventional narrative about leadership, responsibility, and the charismatic CEO seems to be being rewritten chapter by chapter as algorithmic governance develops in real time, and no one has yet seen see how it will finish.
