A Economist's Data Challenge to AI Security Panic
Cybersecurity stocks have been volatile in recent weeks, and both sides of the AI safety debate have claimed the swings validate their position. Callum Williams, an economist, published an analysis arguing that the market movements, when compared to historical norms, do not signal that anything fundamental has changed about how investors view AI-related risk. The data suggests the recent volatility is noise, not signal.
Williams posted his analysis on Marginal Revolution, the economics blog run by Tyler Cowen and Alex Tabarrok. The post includes a graph tracking cyber stock performance over several years. The key finding: cyber firm share prices moved far more dramatically during 2020 to 2022 (when they rose) and during 2022 to 2023 (when they fell) than they have in recent weeks. Yet nobody interpreted those larger swings as evidence that AI posed an existential threat.
What the Market Data Actually Shows
The core of Williams' argument is straightforward. If investors believed AI represented a serious existential risk, you would expect to see that reflected in the prices of companies directly affected by AI security concerns. Cybersecurity firms sit at the intersection of AI capability and AI safety. They build the tools that protect systems from AI-powered attacks and develop the infrastructure needed to secure AI deployments.
But the recent price movements in these stocks fall within normal historical volatility. The 2020 to 2022 rally in cyber stocks was driven by pandemic-era digital transformation and remote work security needs. The 2022 to 2023 decline reflected rising interest rates and a broader tech sector correction. Both moves dwarf the recent swings that some commentators have attributed to AI risk concerns.
Williams' point is that markets process information continuously. If AI existential risk were a serious near-term threat, cybersecurity stocks would likely show sustained upward movement as investors priced in the need for defensive tools. Instead, the stocks have fluctuated within ranges that look unremarkable against the multi-year backdrop.
Data Versus Narrative in the AI Safety Debate
The analysis arrives at a moment when the AI safety discourse has become increasingly heated. Public figures, researchers, and policymakers disagree sharply about the timeline and severity of AI risks. Some warn that advanced AI systems could pose catastrophic threats within years. Others argue those fears are overblown and distract from present-day harms.
Williams' intervention is notable because it grounds the debate in market data rather than theoretical arguments. Markets aggregate the beliefs of millions of investors with real money at stake. They are not infallible, but they represent one of the most concrete signals available about collective expectations.
Tyler Cowen, who published the post, noted that he personally remains more pessimistic than the numbers indicate. But he argued the analysis represents progress because it moves the conversation toward concrete, measurable evidence rather than speculation. His framing was pointed: you should ask who insists on looking at data, and who tries to steer you away from it.
Implications for Security Teams and Investors
For developers and security teams building AI systems, the market data has practical implications. If the market does not price in elevated existential risk, it suggests that current investment patterns in AI security infrastructure reflect normal business cycle dynamics rather than emergency preparedness.
This does not mean AI security is unimportant. The opposite is true: cybersecurity firms continue to invest heavily in AI-powered defense tools, and the demand for AI security expertise remains strong. But the investment thesis is grounded in conventional threat models, not doomsday scenarios.
For investors, the analysis suggests that recent volatility in cyber stocks presents opportunities or risks based on fundamental analysis rather than narrative-driven trading. The stocks that rose on pandemic-era demand and fell on rate hikes are now trading based on their actual business performance and growth prospects.
A Call for Evidence-Based Discussion
Williams' analysis is not the final word on AI risk. Markets can be wrong, and the absence of a price signal does not prove the absence of danger. But it does shift the burden of proof. Those claiming that AI poses an existential threat must explain why the people with the most financial exposure to that scenario are not pricing it in.
The broader lesson for the tech community is that data should anchor discussions about AI risk. Theoretical arguments and thought experiments have their place, but they should be tested against observable evidence. Cybersecurity stock prices are one such test, and right now they are not confirming the alarmist narrative.
This does not resolve the debate. But it gives everyone a concrete starting point, which is more than most recent contributions to the discourse have provided.