Zero-Day Market Economics in 2026: How AI Vulnerability Discovery Changes the Price of Exploits

A zero-day is a vulnerability unknown to the vendor and for which no patch exists. The zero-day market is the ecosystem where these vulnerabilities are bought and sold before vendors are aware of them. The market has three tiers. The top tier is nation-state acquisition through brokers or direct purchase: government intelligence and military buyers paying premium prices for exclusive access to high-value targets like mobile operating systems, encrypted messaging apps, and network infrastructure. The middle tier is defense contractor and private intelligence firm acquisition for vetted government clients. The bottom tier is public bug bounty programs, where vendors pay researchers to find bugs in their own software, typically at prices far below market rates for offensive-grade research.

Zerodium (Washington DC, founded 2015 by Chaouki Bekrar) is the most transparent broker, publishing a public acquisition price schedule. Crowdfense (Abu Dhabi) operates a similar model serving government clients globally. MarketplacePWN, OffSec, and dozens of less visible entities operate in the same space. The dominant buyers are signals intelligence agencies (NSA, GCHQ, DGSE, BND, and equivalents), military cyber commands (US CYBERCOM, GCHQ NCSC, Unit 8200), and defense contractors acting as acquisition intermediaries. Corporate espionage buyers exist but represent a small fraction of market volume by value.

Zerodium's published schedule (updated periodically) reflects relative exploit difficulty and target value. iOS full chain remote jailbreak: up to $2.5 million. Android zero-click full chain: up to $500,000. Chrome RCE plus sandbox escape: up to $250,000. WhatsApp or Signal zero-click: up to $1.5 million. Windows LPE: up to $200,000. VMM escape: historically $100,000 to $300,000. Network equipment (routers, firewalls) varies widely. Nation-state direct acquisition prices are reported by journalists and researchers to be 3 to 10 times higher than published broker rates for the most sensitive capabilities, because exclusive acquisition and operational security have additional value.

Zero-day prices are high because supply is low. Finding a remotely exploitable zero-day in a hardened target requires expertise in reverse engineering, vulnerability research, exploit development, and target-specific knowledge. A competent exploit researcher might produce two to four weaponizable zero-days per year. The total global supply of high-value zero-days is limited by the number of researchers with the skills to find them. Demand from government buyers is relatively inelastic: intelligence agencies have mandated capability requirements and significant budgets. The result is a sellers market with prices reflecting scarcity.

Claude Mythos scored 21 of 41 on ExploitBench's V8 ACE challenges. Every other tool scored zero. On ExploitGym, Mythos outperforms alternatives by 10.5 times. On SCONE-Bench, Mythos identified findings with an estimated aggregate value of $35 million. These benchmarks measure AI performance on synthetic vulnerability discovery tasks, not on live production targets. But they demonstrate that AI is approaching the capability threshold for autonomous vulnerability discovery in complex, real-world codebases. If AI tools can generate a meaningful fraction of the research output currently produced by elite human researchers, supply increases and prices face downward pressure.

The empirical evidence is early and fragmented. Brokers are not transparent about volume or pricing trends. Researchers who track the market report anecdotal evidence of increased supply in specific categories, particularly browser and JavaScript engine vulnerabilities, where AI-assisted research is most advanced. Whether this is translating to price compression is unclear: the final steps of weaponization and operational integration remain human-intensive, which means AI-discovered vulnerability information still requires significant expert work to become an operational exploit. The market impact is likely lagged, not immediate.

If AI reduces the expertise bar for vulnerability discovery, it changes the offensive capability balance between nations with large pools of elite researchers and those without. Historically, sophisticated offensive cyber programs required years of investment in training highly specialized personnel. AI-assisted vulnerability research could allow nations with smaller technical workforces to develop meaningful offensive capabilities more rapidly. This is the democratization of offensive capability argument: the same dynamic that Project Glasswing represents for defense (AI enabling a small team to find what previously required hundreds) applies symmetrically to offense.

The strongest form of the argument is: if Claude Mythos can find 21 V8 ACEs autonomously, what prevents an adversary from building a similar system and generating offensive capability at scale? The answer involves several friction points. Training a capable AI security system requires data, compute, and alignment work that limits who can realistically do it. Weaponizing a discovered vulnerability into a reliable operational exploit still requires human expertise. And a system generating many low-quality findings is less useful than a smaller set of reliable, weaponizable bugs. The democratization concern is real but the capability gap between discovery and operational exploitation remains significant.

The Wassenaar Arrangement is a multilateral export control regime covering conventional weapons and dual-use goods and technologies. In 2013, Wassenaar added controls on intrusion software, surveillance tools, and IP network surveillance systems. The controls on vulnerability research tools are contentious because they could restrict defensive security research alongside offensive tool export. The US Bureau of Industry and Security (BIS) has worked through multiple revision cycles trying to scope the controls narrowly enough to permit legitimate defensive and research use while restricting weaponized exploit export to adversarial nations. AI vulnerability discovery tools sit in an unresolved regulatory gray zone: the policy apparatus has not yet defined how Claude Mythos-class tools fit into existing export control frameworks.

If AI increases the supply of zero-days and compresses the expertise requirement for finding them, the defender implication is that the window between vulnerability creation and discovery narrows. Vulnerabilities that might have gone undiscovered for years because finding them required rare expertise will be found faster, either by defenders through programs like Glasswing, or by adversaries through equivalent offensive AI development. This argues for continuous AI-assisted vulnerability discovery as a defensive posture rather than periodic manual assessments, and for treating the attack surface as dynamically exposed rather than statically bounded by known vulnerabilities.

The Mythos Brief includes a comprehensive zero-day market pricing reference covering all major target categories, broker comparison, and an analysis of how AI-scale discovery is likely to affect pricing over the next 24 months. It also includes a defensive budget implications model: how to translate zero-day market dynamics into security investment prioritization.