Human hackers still define bug bounty in the AI era

At a glance

What this is: This is an opinion-led analysis that argues AI should augment, not replace, human bug bounty researchers, especially for novel and chained attack discovery.

Why it matters: It matters to IAM and security teams because the same human-versus-automation trade-off is now shaping how they assess identity logic, access chains, and AI agent behaviour in real programs.

👉 Read INTIGRITI's insights on how AI is changing bug bounty and human research

Context

AI-assisted security testing is changing the economics of vulnerability discovery, but it does not erase the value of human judgment. The core governance gap is that automated tooling is strongest at breadth, while attackers and researchers still create value through context, chaining, and adaptation.

For practitioners, that means the question is not whether AI belongs in security programmes, but where it can safely absorb repetitive work without flattening the human insight that finds high-impact issues. In bug bounty and adjacent identity-heavy work, that balance is especially important because access logic, privilege boundaries, and AI agent behaviour often fail in ways scanners do not model well.

Key questions

Q: How should security teams combine AI with traditional AppSec scanning?

A: Use AI to improve triage, summarisation, and rule tuning while keeping deterministic SAST responsible for known vulnerability detection. That approach preserves reliability, reduces noise, and helps developers act on findings faster. The goal is not to replace static analysis, but to make it more usable and better aligned with how engineers work.

Q: Why do human testers still matter in AI-assisted security programmes?

A: Human testers still matter because the most valuable findings often depend on understanding how a product behaves, not just what it contains. AI is good at patterns and known classes of issues, but it is weaker at reasoning across requests, adapting to surprises, and uncovering attack paths that emerge from workflow logic.

Q: What do teams get wrong about AI replacing vulnerability researchers?

A: They assume faster scanning equals complete assurance. In practice, AI reduces the cost of breadth, but it does not eliminate the need for human judgment in chained exploitation, business logic abuse, or emergent systems such as AI agents. The right model is augmentation, not substitution.

Q: How should security teams govern AI agents used for offensive testing?

A: Treat offensive AI agents as distinct workloads with explicit ownership, scoped tools, and logged approvals. Give them only the environments, credentials, and actions needed for authorised testing. Separate research targets from production systems, and review retries, data access, and output handling as part of standard governance, not as an afterthought.

Technical breakdown

Why AI is strong at breadth but weak at attack context

Large language models and other AI tools are effective at repetitive analysis tasks such as surface mapping, known vulnerability classification, dependency review, and configuration drift detection. They struggle when the task requires maintaining state across requests, interacting with live systems, or adapting to unexpected application behaviour. That limitation matters because many material issues emerge only when an attacker reasons about workflow, trust, and state transitions rather than isolated flaws.

Practical implication: use AI for repeatable discovery, but reserve human researchers for context-heavy testing and chained attack analysis.

Human researchers still matter for business logic and AI agents

The article’s central technical claim is that the highest-value findings increasingly sit outside the reach of pattern matching. Business logic flaws, multi-step exploitation paths, and novel attack classes in emerging systems such as AI agents depend on understanding how a product behaves in practice, not just how it is coded. That makes human creativity a control surface in its own right, especially where systems delegate decisions or combine multiple privileges.

Practical implication: prioritise manual testing for workflows, delegation chains, and AI agent interactions that could bypass static detection.

Hybrid offensive testing is becoming the default operating model

A mature programme now combines continuous AI scanning for breadth with time-bound human research for depth. The point is not tool substitution but layered assurance: machine assistance handles the noisy baseline, while human testers focus on what scanners cannot reason about. That same model maps well to identity-heavy environments where standing access, privilege chains, and cross-system dependencies create hidden failure modes.

Practical implication: design testing programmes so AI and human reviewers cover different failure modes instead of duplicating the same checks.

NHI Mgmt Group analysis

Human creativity remains the deciding factor when attack paths depend on context. AI can accelerate discovery, but it cannot fully replicate the reasoning required to connect business logic, identity boundaries, and multi-system trust relationships. That means the organisations that over-automate offensive testing will still miss the issues that matter most. Practitioners should treat human-led research as a core assurance layer, not a legacy indulgence.

AI is commoditising breadth, not replacing depth. Surface mapping and known-pattern detection are becoming table stakes, which lowers the value of undifferentiated scanning but increases the premium on adversarial reasoning. This is especially relevant where identity and access decisions are embedded in workflows, because the most dangerous failures emerge when systems trust a sequence of actions rather than a single credential. Practitioners should expect budget pressure to shift toward hybrid testing models.

AI agent behaviour belongs in the same governance conversation as vulnerability discovery. The article’s mention of emerging AI agents is important because an AI system that can select actions and tools becomes a new kind of security subject, not just another application. That creates a governance overlap between application security, IAM, and NHI-style oversight for delegated access. Practitioners should prepare for testing programmes that evaluate both code flaws and the identities of the systems acting on the code.

Human-shaped security programmes will outperform tool-only programmes. The most resilient operating model is not AI versus people, but orchestration between them. That framing aligns with NIST CSF thinking around continuous improvement and with identity governance principles that separate routine control from exception handling. Practitioners should build programmes where automation raises coverage and humans own the exceptions, the edge cases, and the novel attack paths.

What this signals

Human-shaped assurance is becoming a programme design choice. Security teams that keep treating AI as a replacement for researchers will overinvest in breadth and underinvest in judgment. The better model is to let automation handle repeatability while humans own exception handling, because that is where material risk still concentrates.

Identity and AI governance are converging at the point of delegation. As AI agents take on more tool use and decision-making, their access patterns start to resemble non-human identities that need explicit scope, review, and containment. Teams should align testing, IAM, and security engineering around that overlap before it becomes an operational blind spot.

For practitioners

• Define separate test lanes for breadth and depth Use AI-driven scanning for known patterns, dependency exposure, and baseline recon, then assign human researchers to business logic, chained abuse, and live-system interaction.
• Map AI agent interactions into security testing scope Treat AI agents as operational systems that can make decisions, call tools, and traverse trust boundaries, then test their delegated actions as part of application and identity reviews.
• Prioritise manual review for identity-heavy workflows Put human testers on authentication handoffs, role transitions, session boundaries, and privilege escalation paths where a scanner can confirm symptoms but not exploitability.
• Review programme economics against attack value Compare the cost of continuous automation with the findings produced by human-led deep dives, especially where low-volume issues could have high blast radius if missed.

Key takeaways

• AI is strongest at repetitive discovery, but human researchers still find the issues that depend on context and adaptation.
• The main programme shift is from tool substitution to hybrid assurance, where automation expands coverage and humans handle depth.
• As AI agents gain operational power, their delegated access becomes an identity governance problem as much as an application security one.

Key terms

• Business logic vulnerability: A business logic vulnerability is a flaw in how an application’s workflow or rules are enforced, allowing an attacker to misuse a process rather than break code directly. These issues often evade signature-based tools because the weakness lies in authorisation, sequence, or state handling.
• Hybrid Offensive Testing: A security testing model that combines automated scanning with human-led research. Automation handles repetitive discovery, while humans focus on context, chaining, and emergent behaviour that tools cannot reliably reason about.
• Delegated Access: Delegated access is permission granted to one identity to act on behalf of another user, service, or system. In NHI environments, this usually appears in OAuth-connected apps and automation tooling. It is powerful, but it must be tightly scoped and reviewed because it can persist long after the original business need ends.
• Human-Shaped Security: An operating model that assumes people remain essential to finding, interpreting, and prioritising complex security issues. The term describes programmes that use automation for scale but preserve human judgment for exceptions, novel paths, and high-impact decisions.

What's in the full article

INTIGRITI's full article covers the strategic detail this post intentionally leaves at a higher level:

• The CEO's full argument on why human hackers remain central to bug bounty economics.
• The article's broader view on how AI changes the cost structure of offensive security.
• Additional commentary on where AI helps and where it still falls short in real testing.
• Links to related Intigriti perspectives on AI, triage, and vulnerability discovery trends.

👉 The full INTIGRITI article expands on the role of human hackers, AI limits, and the future of vulnerability discovery.

Deepen your knowledge

The NHI Foundation Level course, the industry's only accredited NHI security programme, covers NHI governance, workload identity, secrets management, and agentic AI identity. It helps security practitioners connect identity controls to the broader risk patterns now emerging in AI-enabled environments.