The first report arrived in 13 minutes. Somewhere outside the Pentagon, a hacker had found a weakness on a public-facing Defense Department website, followed the rules printed for an unusual government experiment, and pressed submit. No cloak, no basement montage, no shouted warning across a command center. Just a form, a finding and a clock that made the old procurement logic look slow.
This was April 2016, and the experiment was Hack the Pentagon, the federal government’s first bug bounty. Over the pilot, 1,410 people registered, 1,189 reports came in and 138 were judged valid and unique. The entire program cost $150,000, roughly half of it paid to researchers. The Defense Department estimated that a comparable conventional security assessment could have cost more than $1 million.
HackerOne hosted the challenge. But the company’s real contribution was less theatrical than “crowdsource the Pentagon.” It built the plumbing that let an institution say who could test what, which techniques were allowed, how reports would be deduplicated, how severity would be judged and how money would move. The hackers supplied curiosity. HackerOne supplied a legal front door and an orderly waiting room.
The product is permission
HackerOne began in 2012, founded by Alex Rice, Jobert Abma, Michiel Prins and Merijn Terheggen. Rice had run product security at Facebook. Abma and Prins were Dutch ethical hackers who knew the other side of disclosure: discover something serious, then struggle to make a company pay attention without making its lawyers nervous. One company initially brushed off their emails, so they sent a cake. The pastry got a meeting. The problem was suddenly obvious enough to build around.
A vulnerability is technical. Disclosure is social.The insight behind HackerOne’s marketplace
Before a formal program, a well-meaning researcher might email support, message an employee or post publicly after silence. Each path carries risk. The organization cannot separate signal from noise; the researcher cannot tell whether curiosity will be rewarded or prosecuted. HackerOne standardized the exchange. A customer publishes scope and rules. Researchers test inside those boundaries. Valid discoveries earn reputation and, in a bounty program, money. The platform routes reports, handles duplicate claims, supports triage and connects findings to tools engineers already use.
The catalog now has several doors. H1 Response is the intake desk for vulnerability disclosure. H1 Bounty adds financial incentives and a researcher crowd. H1 Pentest provides scoped testing by vetted professionals, with findings delivered while the engagement is still running. H1 AI Red Teaming attacks models, agents and their surrounding applications. H1 Code moves review earlier into source code, an expansion helped by HackerOne’s 2022 acquisition of PullRequest. Hai, its AI layer, queries program data, adds context and helps with triage and remediation.
The Pentagon changed the buyer
Bug bounties were not invented by HackerOne. Technology companies had paid outsiders for useful flaws for years. What changed with the Pentagon was the buyer’s mental model. The organization most associated with locked doors had demonstrated that permission could be safer than silence. The pilot’s first failure was not a system outage or a catastrophic breach. It was noise: 1,189 reports narrowed to 138 valid, unique issues. That 88 percent reduction is precisely why triage matters.
Two ways to buy a security test
The numbers changed minds because they were legible. A first report in minutes. A finite budget. Valid findings delivered to people who could fix them. The Defense Department followed the pilot with disclosure policy work and additional military challenges. For enterprise buyers, HackerOne gained a reference customer that made the practice easier to defend in a boardroom. If the Pentagon could publish a scope and invite scrutiny, a bank or retailer had less reason to pretend outsiders were not already looking.
Today the customer list crosses Anthropic, Adobe, Coinbase, General Motors, Goldman Sachs, GitHub, Hyatt, Lufthansa, PayPal, Uber and both the U.K. and U.S. defense establishments. HackerOne says more than 1,300 companies trust the platform and more than 600,000 bugs have been found. Its 2025 report covered more than 580,000 validated vulnerabilities and nearly 2,000 enterprise programs active during the year. Researchers received $81 million in payouts.
A market for attention
HackerOne is an enterprise software company wrapped around a two-sided marketplace. Customers generally buy a platform subscription and managed services, then fund rewards for valid reports. Pentests and red-team exercises are scoped engagements. Prices are not posted publicly, which makes the cost less tidy than the Pentagon anecdote. The total depends on program scope, service level and how many valuable vulnerabilities researchers find.
The marketplace works because researcher attention is scarce. A good hunter can choose among programs and will weigh payout, scope, response speed and the chance of filing a duplicate. HackerOne’s data contains a blunt lesson: when programs reduce rewards, valid and critical submissions often fall too. Calling people a “community” does not suspend economics. If a company wants sustained, high-quality attention, it must make the work worth doing and respond like the work matters.
AI creates bugs - and paperwork
The company’s newest opportunity is also a threat to its original model. AI writes more code, changes products faster and gives attackers leverage. It also lets inexperienced researchers produce floods of plausible-looking reports. More discovery is not automatically more security; it can be a larger inbox. HackerOne says vulnerability submissions rose 92 percent year over year ahead of its 2026 H1 Platform launch, while remediation throughput lagged.
Its answer is a human-machine pairing. AI agents can perform reconnaissance, repeat tests, enrich reports and draft fixes. Human researchers still decide whether a flaw matters in a real business, connect weak signals into an attack chain and exercise judgment around safety. In HackerOne’s 2025 research, 58 percent of surveyed researchers said AI misses business logic or chained exploits, while only 12 percent thought it could replace them.
AI is excellent at making more findings. Security is the art of deciding which finding can become a bad afternoon.The discovery-remediation gap
The attack surface has expanded beyond conventional software. HackerOne researchers now probe prompt injection, tool misuse, context poisoning, data leakage and unsafe model behavior. During an Anthropic challenge, researchers spent more than 3,700 hours and generated over 300,000 interactions against a demo protected by Constitutional Classifiers. They found no universal jailbreak, but the exercise informed classifier tuning and refusal thresholds. The outcome was not a cinematic breach. It was evidence about where a defense held and where it needed adjustment.
The copyable part is boring
A company does not need two million registered hackers on day one. It needs a clear disclosure policy. Name the assets researchers may touch. Grant safe harbor for good-faith work. State what is out of bounds. Acknowledge reports quickly. Put an engineer, not a generic support queue, behind the inbox. Define severity before money is involved. Track time to triage, time to remediation, duplicate rate and reopen rate. Start privately if the team needs practice, then widen the invitation.
The approach is a poor fit when those operational muscles are missing. An open bounty can overwhelm a small team, expose ambiguity in asset ownership and create a reward bill that finance did not expect. It will also disappoint if leaders treat the program as proof that the product is secure, or if engineers cannot fix what arrives. A scoped pentest may be better for a new system, a compliance deadline or a team that needs predictable effort. Automated scanners remain useful for cheap, repeatable checks. The models complement each other.
That is where HackerOne sits in the market: between scanners that know patterns, consultancies that sell defined expertise and crowds that supply diverse adversarial attention. Its defensibility comes from coordinating all three, plus more than a decade of vulnerability data and trust accumulated on both sides of the report form. Competitors can offer similar mechanics. The harder thing to reproduce is a marketplace where skilled researchers believe their work will be judged fairly and large institutions believe outsiders can be let in safely.
Thirteen minutes after the Pentagon opened its door, someone knocked. HackerOne’s bet is that the knock was never the danger. The danger was having no door, no rules and nobody listening on the other side.