HackerOne joins human security research with agentic testing and remediation

HackerOne combines vulnerability programs, Hai agents and code-informed remediation, with separate platform, credit and reward budgets.

01 / ProductAI agents operate alongside a security-research program

HackerOne connects organisations with security researchers and supplies the platform for receiving, validating and acting on vulnerability reports. Its current H1 offer adds agentic testing and remediation alongside bounty, disclosure and assessment programs. Hai is the company’s AI system, with agents aimed at tasks such as improving report completeness, finding duplicates, escalating priorities and adding historical context. These tasks support a security workflow rather than replacing its accountable owners.

H1 Continuous Testing describes specialised agents that map an application and test for exploitable weaknesses as it changes. H1 Remediation works from validated findings to produce a code-informed fix plan. Discovery and remediation are different stages: a finding needs enough evidence to be actionable, while a proposed correction still needs engineering review, deployment and retesting before the exposure is resolved.

The company overview retains HackerOne as the identity behind these offers. Its acquisition of PullRequest also provides context for its code-review capabilities; it does not create another company entry here. Sequenced reviewed public product and commercial sources, without running an assessment or validating HackerOne’s advertised accuracy and risk-reduction metrics.

02 / AudienceThe strongest fit is a team that can act on external findings

An application-security group may already receive useful reports but lose time turning them into reproducible engineering tasks. Reports vary in completeness, several researchers may describe the same root cause and business impact can be unclear to the service owner. HackerOne is relevant when the team wants a coordinated intake and validation process with enough context to prioritise the actual work.

A rapidly changing application presents another case. Periodic assessments can leave long intervals between changes and testing. An agentic testing offer is worth evaluating when the organisation can define authorised scope, supply representative access and review findings continuously. It is not a substitute for ownership. A larger stream of validated issues still needs engineers, deployment windows and someone responsible for accepting or resolving risk.

Compare GitLab when the primary decision concerns security checks within the software delivery platform. Compare Torq when the challenge is coordinating security actions across tools. HackerOne’s distinctive starting point is a vulnerability program that brings researcher and testing evidence into that workflow. The comparison should follow the source of the findings and who acts on them, rather than placing every security agent in the same category.

03 / WorkflowProposed workflow: carry one validated finding through to a verified fix

Select an application the organisation is authorised to test and define a bounded scope with its owner. Use a staging environment where possible, with test accounts and data. Document excluded paths, permitted testing intensity and an escalation contact. This proposed pilot is an evaluation of the report-to-remediation process; it is not an invitation to test unrelated systems or a description of work already performed.

Begin with one existing validated report and inspect whether the affected endpoint, preconditions, reproduction steps and impact are clear. Use Hai’s documented report and context assistance to prepare a consistent case. Keep original researcher evidence accessible. If an agent proposes that two reports are duplicates, have the reviewer compare the root cause and affected scope before closing one. Similar symptoms can come from different defects.

For continuous testing, map the approved application and its authentication boundaries first. HackerOne describes reconnaissance before testing and incremental checks around changes. Evaluate that claim using a controlled application change whose security implications are understood. The desired output is a reproducible finding tied to the tested version, rather than an unbounded list of possible weaknesses with no proof that they exist in the environment.

Connect source code only where the team has approved the access and the feature is included. The remediation product describes repository-informed root-cause analysis and delivery of fix plans to development tools or an AI coding agent through MCP. Review what repository, branch and code region the plan actually references. A finding from a deployed version can be misinterpreted if analysis silently uses a different revision.

Have the service owner inspect the proposed fix and implement it through the normal change process. Include tests that reproduce the original issue and demonstrate the expected authorised behavior after correction. Keep the test within the agreed scope. Neither a generated patch nor a closed ticket proves that the deployed service changed; verify the released version and retest the originally demonstrated weakness.

Close the loop with the researcher or testing program using the organisation’s established communication process. Preserve the reason for the final disposition and any remaining limitations. Measure the time from a validated report to a verified deployment, the number of reports reopened and the effort spent clarifying findings. These metrics are more useful than simply counting generated fix plans, because they reflect the actual end of exposure.

04 / PricingPlatform subscription, agentic consumption and rewards are different budgets

The H1 pricing page describes Professional, Enterprise and Enterprise Plus platform editions. It presents a committed subscription plus prepaid credits and separately controlled researcher rewards. The page does not publish a universal currency price. Professional lists disclosure coverage, asset inventory, Hai Chat, signal enrichment and governance features; higher editions extend governance and scale. Confirm the selected edition and activated products in the proposal. Its FAQ also states that bounty programs carry a rewards service fee, with the fee confirmed in the quote; the reward pool is therefore not the entire bounty-program cost.

The agentic-credit documentation describes consumption for Asset Intelligence, Continuous Testing and Remediation. Within Asset Intelligence, only Intelligent Scoping Recommendations are identified as credit-chargeable. Administrators control activation, and certain products require signed terms before their switches become available. Hai acts as the master enablement switch, so access to the assistant should not be interpreted as unlimited activation of every agentic product.

The default exhaustion behavior is unusually consequential: no further agentic runs start after credits are depleted, while a run already underway completes. Overage is disabled by default and requires an arrangement with the customer-success team. A testing program that assumes continuous coverage should therefore monitor its balance and decide what happens when the allowance runs out. Researcher reward budgets need their own owner and approval path.

05 / DifferenceExploit evidence can travel into the engineering task

HackerOne’s valuable distinction is the opportunity to connect a demonstrated weakness to its source-level cause and remediation plan. The finding carries context about preconditions and impact that an engineering team might otherwise reconstruct from a scanner result. That context can help a developer choose a correction that addresses the actual exposure instead of merely suppressing a symptom.

The human research component remains relevant as AI-generated software changes application behavior. Business logic, trust relationships and unusual workflows can be difficult to describe with a generic rule. The platform’s combined approach is worth evaluating where human findings and repeatable agentic checks inform one another. This is an architectural and workflow rationale, not evidence that the company’s agents or researchers will discover every issue.

06 / LimitsTesting scope and product terms are part of the implementation

Ask which authenticated workflows the testing product can reach and how it handles state-changing actions. A product that can map public endpoints may still need configuration to exercise tenant-specific permissions or a multistep business process. Include those limitations in the coverage record. Absence of a finding only describes what the authorised test actually examined, with the access and configuration it had.

Public packaging also evolves. The current pricing page names more credit-based capabilities than the help article’s three-product overview. Treat the article as guidance for the products it explicitly documents, and obtain the current menu, activation terms and consumption rules for any additional purchase. Do not infer a universal credit rate or stopping rule for a newly named capability solely from older documentation.

For code-informed remediation, settle source access and the expected output. A plan delivered to Jira or an AI coding agent is not necessarily a merged pull request, a passing build or a production release. Retain normal engineering review and regression tests. Also decide how the company will communicate a disputed finding, a partial fix or a retest that still succeeds; those cases determine whether the workflow remains credible under pressure.

07 / DecisionEvaluate risk resolution, not just faster report processing

HackerOne is a strong candidate for teams that want vulnerability research, testing and remediation to form one accountable process. Start with a narrow authorised application, carry a validated issue through engineering and verify the deployed fix. Then size platform access, agentic credits and researcher rewards separately. The useful result is a repeatable reduction in demonstrated exposure with evidence that both security and engineering can inspect.