The Economics of Security Vulnerabilities: Why Discovery Is Not Commoditizing

Asking a frontier model to find a vulnerability costs pennies, yet black hat brokers are paying up to $9 million for a working zero-click exploit chain against a modern smartphone, and Chrome and Safari exploits trade at $3 million.

If discovery is cheap, why are people willing to pay more for exploits than ever before?

In recent months I’ve been hearing this narrative: Since AI finds bugs and everyone has access to AI, discovery is becoming a commodity and the interesting work has moved to triage, remediation, or some other “higher-value” tasks.

But the price signals are unmistakable: discovery is not really commoditizing.

What the Market Shows

There are a few areas where things really have become cheap and way more accessible. Both shallow discovery (especially in code that never received serious scrutiny) and exploit reproduction for already disclosed vulnerabilities are now easy to automate and have lost most of their value.

But this is not commoditization. Plenty of discoveries and reproductions still resist serious effort, even from frontier AI, and token usage at scale can get very expensive very fast. Plus, consider the operational complexity of incorporating detection and triage into actual workflows. In replication and discovery alike, the floor has moved up, but it is still very far from the ceiling.

Think about what this means for the price of exploits. When everyone queries the same handful of public models, discovery becomes correlated. Whatever a generic prompt surfaces for you it surfaces for every other user of that model, so the researcher, the defender, and the attacker are all drawing from one shared brain.

In other words, the exclusivity window of a one-shot finding is pretty much zero, because exclusivity is the entire source of a vulnerability's value. An exploitable flaw is worth millions while few know about it, and it approaches zero the moment everyone does.

So one-shot discovery did not commoditize the valuable tail. It redefined the bulk, while making the tail even more valuable. The floor of what counts as a trivially findable bug rose significantly, which is real progress, and exactly why the field feels commoditized from the inside. But it doesn’t capture the shape of the whole problem.

The market's verdict is in the prices. If discovery were commoditizing, the new supply would be pushing down the premiums. But instead, this is what happened: a decade ago, brokers paid tens of thousands of dollars for serious exploits. In 2016, the top bounty was $1.5 million. By 2019, $2.5 million. Today the top of the market sits between $7 and $9 million. And this steep increase has happened through exactly the same period in which fuzzing, SAST, formal verification, and now one-shot AI discovery raised the floor.

Nicole Perlroth's excellent This Is How They Tell Me the World Ends traces this market back to its origins, when brokers paid a few hundred dollars per bug. In other words, the longest price series we have for exclusive discovery points in one direction only. Markets do not pay escalating premiums for commodities, they pay them for specialty goods.

Why the Exploit Premium Endures

Two properties of vulnerabilities keep exploits valuable.

First, their value distribution is a power law. The finding that every tool surfaces is worth roughly nothing, because it is defended everywhere within days. On the flip side, the critical finding only one engine surfaces is worth everything.

A commodity capability, by definition, finds what everyone finds, which is the worthless bulk. But there are many important things outside of that bulk. We see this in our own comparative work: different analysis engines running against identical codebases produce finding sets that barely overlap. Buyers see it too. A security team recently told us they had evaluated several AI code analysis vendors alongside their own harness built on foundation models, and they found vulnerabilities that every vendor missed, while the vendors found ones their harness missed. Their conclusion? No approach is definitive.

That observation is correct, and it shows that the market is far from commoditized. Commodities produce interchangeable output, but discovery engines produce different slices of an unfinished map.

Second, and more importantly, the tail moves. "Beyond current tooling" is a moving boundary, not a fixed niche. Every new AI model or capability increment may raise the floor, but it also converts yesterday's undiscoverable exploit into today's ceiling and exposes a new tail behind it. This is why fuzzing never "finished" OpenSSL, why a decade of continuous automated analysis on the same targets keeps producing findings, and why broker prices rise despite tooling improvements.

The market is simply doing what markets always do, which is repricing whatever remains scarce. Even though the overall number of security bugs is clearly finite, we’re nowhere near the saturation point. We’re only limited by our current toolset.

The Apex Is in Your Stack, Not in Cupertino

The apex vulnerability is different for each target. An iOS exploit chain can be priced at $9 million because it is very difficult to find, so discovery is brutally expensive. But the apex for a typical enterprise is an unauthenticated remote code execution flaw in a VPN appliance, a file transfer tool, or an email gateway. Unsurprisingly, that is exactly where the threat actor economy concentrates, because value scales with deployment count and the attractiveness of the target.

Consider what the most damaging vulnerabilities of the last decade had in common. MOVEit was an ordinary bug class in unglamorous enterprise code, and it breached thousands of organizations. Log4Shell was a documented convenience feature in a logging library running on hundreds of millions of systems, repurposed into remote code execution. Heartbleed was a trivial buffer over-read that survived years inside one of the most scanned codebases in existence. Nothing about these bugs was technically spectacular. What made them apex exploits was their location, reachability, and above all, their exclusivity.

And when exclusivity gets extreme, it looks like XZ-Utils: a backdoor planted inside a compression library on its way into every major Linux distribution, known to exactly one party. Importantly, it wasn’t discovered by any scanner, but because one engineer got curious about half a second of login latency. That is what the tail of the distribution looks like in practice.

The Economics of Defense

Now consider what a broker price really is. A rational buyer paying $9 million expects to extract more than $9 million from their targets. In the case of MOVEit, the losses across all victims ran into the billions of dollars. So if anything, the price of exclusive discovery is an extremely conservative measurement of expected losses.

Defenders have completely different budget constraints than attackers, but underground broker pricing is still the best way to evaluate their expected value. After all, compliance and labor metrics only price the process. To estimate the risk, you need to look at attacker economics.

By implication, a security tool can look great in terms of compliance and labor-saving, but leave expected losses untouched because it doesn’t provide exclusive discovery, which is where risk lives.

The Capability Arms Race Is On

In normal markets, the median buyer defines the quality bar. In adversarial markets, the best-resourced adversary does. That’s why raising the floor doesn’t make the world effectively more secure, it just sets off an arms race.

And the AI version of this deserves to be stated plainly. If your discovery capability is a prompt against the same public model your adversary uses, your coverage only protects against the attacker's baseline. You find what they find, when they find it. That is not a defense. That is a tie.

Finding Exclusive Vulnerabilities With AISLE

The vulnerabilities that matter most are the ones only one party knows about, and as broker pricing reveals, the ability to find them is getting more valuable with each advance in capability. The question for defenders is not whether your scanning is good enough. It is this: who reaches the tail of your stack first?

Want to see what the industry-leading AI security engine finds in your code? Get a one-time code audit, AISLE Snapshot, in the environment of your choosing: air-gapped, on-prem, or in the cloud. Get your Snapshot.