Technology & AI
8 min read

A Wake-Up Call on Cyber Risk as AI Ups the Stakes

Concentrated tech markets and misaligned incentives are leaving companies and governments exposed to cyberattacks.

Susan Athey Cyber Risk Key Image
A security flaw in widely used software can affect millions of users at once. | iStock/JuSun

November 16, 2026

| by Dave Gilson

In Brief

  • Organizations that rely on the same software are vulnerable to simultaneous cyberattacks, especially as hackers use AI to find flaws.
  • Many firms underinvest in cybersecurity, expecting competitors, the government, or IT providers to protect shared tech infrastructure.
  • Better information sharing, public investment in cyberdefense, and rethinking insurance could help realign the incentives for mitigating cyber risk.

In October, security researchers announced that they’d used AI to uncover security flaws in PHP, a programming language that runs on an estimated 70% of the world’s websites. It was a textbook case of how one weakness in a widely used tool can expose millions of users to simultaneous attack.

As more organizations and sectors come to rely on the same critical technology, the threat of systemic cybersecurity incidents is increasing, according to Susan Athey, a professor of economics at Stanford Graduate School of Business, Neil Gandal of Tel Aviv University and the University of Tulsa, and Tyler Moore of the University of Tulsa. The potential fallout goes far beyond the scramble to install patches. “The most severe cyber incidents,” they write, “can lead to losses amounting to billions of dollars and also have the potential to result in physical harm, loss of life, or even political turmoil.”

In a new article in the Journal of Economic Perspectives, the three economists lay out a sobering picture of the cybersecurity landscape and the economic factors standing in the way of improving it. Large-scale cyber risk is a well-known problem. But Athey, Gandal, and Moore argue that the incentives for mitigating it are misaligned.

Tech product markets tend to be concentrated due to economies of scale: Bigger is better. That means when a single software firm serves a large share of its product category, a security flaw can affect many of the customers in that market. However, software makers are rarely held liable for the downstream damage caused by security gaps in their products.

With her coauthors, Athey, a senior fellow at the Stanford Institute for Economic Policy Research (SIEPR), asks: Do market forces discipline cybersecurity? Not very well, her team finds. Companies and other organizations often can’t judge how secure software products are before buying them, and switching vendors or platforms is expensive, even after a security incident. When breaches cause widespread disruption, customers tend not to punish individual organizations. As a result, Athey and her coauthors write, “cybersecurity investment is below the social optimum” for organizations and the software vendors that serve them.

A lack of investment in cybersecurity also imposes costs on residents who depend on local governments for critical services. For example, municipalities in the U.S. typically operate water and wastewater systems, but often lack the scale and incentives to invest in cybersecurity. In July 2026, sophisticated hackers attacked municipal water systems across seven states, locking officials out of the automated control systems that manage pumps, valves, and other infrastructure. Athey, Gandal, and Moore say it is likely that the use of common (and often outdated) infrastructure enabled these simultaneous attacks.

The stakes are rising as AI tools “have advanced to the point where they can act as autonomous agents, searching for the same weaknesses across many enterprises.” (The PHP vulnerability was just one of several security issues uncovered by AI during what some experts are calling “Black October.”) The solution, Athey and her coauthors argue, is for policymakers, regulators, and insurers to step up and reshape the incentives for securing technologies that offer shared benefits — and risks.

The State of Cyber Risk

Data compiled by Athey, Gandal, and Moore show growing threats, concentrated technology, and a fragile safety net. Some of the numbers:

4x: The growth in ransomware attacks between 2021 and 2025. Ransomware, malicious software that locks users out of their files or systems until a ransom is paid, is now “the dominant cybercrime threat to organizations.”

Quote
Firms benefit even if they don’t contribute to preventing cyber risk, so many expect others to bear the costs of protecting shared tech infrastructure.

23%: Share of U.S. municipalities (versus 64% of state governments) with security controls in at least one of the five National Institute of Standards and Technology cybersecurity categories (identify, protect, detect, respond, and recover). Just 5% of municipalities (versus 49% of state governments) have controls in two or more NIST categories; just 2% of municipalities cover three or more.

18,000: The number of organizations in 36 countries that downloaded malicious code hidden in SolarWinds’ network-monitoring software in 2020. Few customers abandoned the product afterward. “Market incentives do not seem to have mitigated the systemic cyber risk created by market concentration,” Athey and her coauthors write.

51%: Cisco’s share of the U.S. router market in 2022. Athey and her coauthors find a strong correlation between a router vendor’s installed base and the number of known vulnerabilities in its products. Cisco routers were the entry point for Salt Typhoon, an intrusion into U.S. telecom networks discovered in 2024 and linked to the Chinese government.

More than 50%: The share of cybersecurity insurance premiums that are passed along to reinsurers (the companies that insure insurers). Reinsurance allows insurers to take on large, complex risks that no single company could absorb, but the cyber reinsurance market is highly concentrated: The five largest players underwrite more than 60% of premiums. “Should any major reinsurer curtail or exit its cyber capacity, the consequences for primary insurers — and for the market as a whole — could be severe,” the authors write.

Fixing the Incentives

Measuring the risks, the authors say, is the first step. The next, more difficult step is to rethink the incentives for reducing cyber risks.

Policymakers could respond by beefing up compliance requirements for firms, as has been done in the European Union. Athey, Gandal, and Moore are wary of this approach. They note that U.S. defense contractors were expected to meet more than 100 security requirements until recently. “No empirical justification was given for why these more than 100 requirements were chosen,” they write. Few firms adopted the rules, and the compliance costs threatened to shut out small, innovative suppliers. A 2025 study coauthored by Gandal and Moore found that the list could be cut by more than half without losing much value. This year, the Pentagon made moves toward a “best practices” standard that could allow new ideas and products to enter the market.

Another possible response to cyber risks would be to hold software companies legally liable for damages caused by attacks on their products. Here, too, Athey and her coauthors are skeptical. Under U.S. tort law, unlimited legal liability “could easily make developing new software, even software intended to fix cyber problems, into an investment that no firm would wish to undertake.”

Instead, the authors suggest mechanisms that promote cybersecurity as a public good. Everyone can benefit from cyber risk prevention, and disclosing new vulnerabilities or promoting new fixes costs very little. But because firms benefit whether or not they contribute to these efforts, many underinvest and expect competitors, the government, or IT providers to bear the costs of protecting shared tech infrastructure. Addressing this “public goods problem” will require new incentives and institutions.

One starting point is filling information gaps. Companies rarely know when their peers have been hacked or which security products actually work. Just as the government provides detailed weather reporting to help farmers, it could publish data on how cybercriminals profit from attacks. To protect trade secrets, firms could share incident data only with each other or with regulators, possibly in exchange for limited liability protection. Athey and her coauthors caution against requiring full transparency, as it could disincentivize security firms from conducting proprietary threat research.

Athey, Gandal, and Moore also propose that governments pay for cyberdefense directly. AI has made it cheaper to find and fix vulnerabilities, and governments could buy and share such tools. The authors see promise in “bug bounty” platforms, which pay researchers for uncovering vulnerabilities before they’re exploited. Requiring that those flaws eventually become public would spread the benefit.

Finally, insurers could do more to understand and reduce the risks faced by their customers. They can improve cybersecurity audits and share lessons from past security incidents. They could also reward preparedness. “Firms with better demonstrated recovery capabilities are lower risk for ransomware payouts and, thus, can be offered lower premiums,” the authors write.

None of these fixes would eliminate the risk of a catastrophic cybersecurity failure, and Athey, Gandal, and Moore note that their essay is “only a stepping stone toward the further research and concrete actions that are needed to address systemic cyber risk.” But with AI agents scouring the internet for the next PHP, they argue, this kind of economic analysis is critical.

For media inquiries, visit the Newsroom.

Explore More