Agentic AI can surface overlooked web flaws faster than manual review

At a glance

What this is: PortSwigger shows that agentic AI can push deeper into large, low-priority code surfaces and reveal web flaws that timeboxed human review often skips.

Why it matters: This matters because pentest coverage is constrained by human time, and identity-related exposure paths such as whistleblower access controls can be missed until an attacker or AI-assisted reviewer finds them.

Context

Timeboxed penetration tests force security teams to triage aggressively, which means large client-side codebases and obscure access paths are often deprioritised even when they contain real exposure. In web testing, the problem is not only finding flaws, but spending enough time on low-signal areas for those flaws to become visible.

This case sits at the intersection of web security, access control, and identity governance. A whistleblower portal is not just an application feature, it is a protected trust boundary, and weak report-code protection turns that boundary into an access problem as much as a code-quality problem.

Key questions

Q: What fails when confidential portals rely on short access codes?

A: Short codes fail when they are used as the main barrier to sensitive content without strong rate limiting, lockouts, and monitoring. At that point the scheme behaves like weak authentication rather than real authorisation, and automated guessing can turn a hidden workflow into a data exposure event.

Q: Why do timeboxed pentests miss important web flaws?

A: Timeboxed pentests miss important flaws because analysts must make hard trade-offs between depth and coverage. Large bundles, obscure workflows, and low-signal code paths are easy to deprioritise, so the engagement can finish with only the most obvious issues confirmed while meaningful exposure remains undiscovered.

Q: What signs show that an application needs AI-assisted review?

A: The strongest signals are oversized client-side bundles, repetitive inspection tasks, and workflows where business logic is likely hidden behind minified code or indirect navigation. Those conditions suggest that human-only review may run out of time before the highest-risk paths are properly analysed.

Q: Should security teams use agentic AI before or after manual validation?

A: Security teams should use agentic AI before manual validation, not instead of it. The AI is best used to widen coverage and surface candidates, while the human tester confirms exploitability, tests business impact, and decides which findings are real enough to escalate.

Technical breakdown

Why large minified JavaScript hides exploitable logic

Minified JavaScript compresses application logic into dense, low-readability bundles that are expensive to inspect manually. Traditional scanners can flag the presence of exposed scripts, but they do not reason about what the code does, how endpoints are wired together, or whether the script reveals protected workflows. Agentic AI can traverse the bundle, infer purpose, and connect routine UI code to sensitive backend functions. That makes it useful for surfacing disclosure paths that would otherwise sit below the human reviewer’s time budget. In practice, the risk is not simply exposed source code, but exposed business logic that points to a reachable attack path.

Practical implication: test large client-side bundles with tools that can reason over code structure, not just detect file exposure.

How weak report-code schemes become brute-forceable access controls

A six-digit alphanumeric code creates a small enough search space that it can become a practical access mechanism problem if rate limiting, lockouts, and monitoring are absent or weak. Once the code is the only barrier to confidential content, the control behaves like a low-entropy authenticator rather than a real authorisation layer. Burp-style automation can then validate whether the portal merely obscures access or actually governs it. The key security issue is that secrecy is being used as a substitute for robust entitlement control.

Practical implication: treat short codes guarding sensitive content as authentication controls and test them for brute-force resistance and abuse detection.

Why agentic AI changes penetration-testing coverage

Agentic AI does not replace the tester’s judgment, but it changes the economics of depth. Instead of spending all available time on obvious defects, an analyst can delegate repetitive inspection and use the returned findings to focus on validation, chaining, and impact. That matters because many important vulnerabilities are not complex in principle, they are simply expensive to uncover at scale. For practitioners, the architectural lesson is that coverage gaps are often a workload problem, not a lack-of-skill problem. AI helps convert analyst time into broader inspection of neglected surfaces.

Practical implication: use agentic AI to extend review depth on low-priority areas, then reserve human effort for confirmation and impact analysis.

Threat narrative

Attacker objective: The attacker wants to retrieve confidential whistleblower reports by bypassing the portal's weak report-code protection.

1. Entry occurs through a client-side JavaScript bundle that exposes enough application logic for a reviewer to infer where confidential whistleblower reports are handled.
2. Credential or access abuse becomes possible because each report is protected only by a six-digit alphanumeric code that can be brute-forced with automation.
3. Impact is confidential whistleblower reports being exfiltrated from the portal, showing that a weak access gate can expose sensitive data at scale.

Breaches seen in the wild

• iOS apps leaking hard-coded secrets: Cybernews found 71% of 156,080 iOS apps leak hard-coded secrets, with open cloud storage and Firebase databases exposing user data.

Read and download The State of NHI & AI Agent Breach Report 2026, covering 200+ breaches impacting Non-Human Identities including AI Agents.

NHI Mgmt Group analysis

Agentic AI is now a coverage multiplier for web testing, not just a novelty layer. The practical change is that pentesters can spend less time on mechanical inspection and more time on validation, chaining, and impact. That shifts the bottleneck from analyst patience to quality of delegated inspection. For practitioners, the security value lies in using AI to expand what gets examined before the engagement clock expires.

Low-entropy access gates are a governance failure when they protect confidential content. A six-digit alphanumeric code is not a durable entitlement control if it can be brute-forced without strong rate limiting and detection. The issue is not only confidentiality, but the assumption that obscurity is equivalent to authorisation. Practitioners should treat these mechanisms as part of access governance, not just application design.

Timeboxing creates a blind spot that attackers can exploit long before the next assessment. This case shows that routine penetration testing can still miss important flaws when the inspection cost is too high for human-only review. The lesson for security programmes is that coverage assurance has to account for neglected code surfaces, not only confirmed findings. Practitioners should reframe pentest success as depth of inspection, not just number of issues found.

Whistleblower portals need stronger identity and access control assumptions than ordinary content delivery. When confidential reports are protected by a short code rather than robust authorisation, the trust boundary is too weak for the sensitivity of the data. This is an identity governance problem as much as a web testing problem, because the control is deciding who can see protected information. Practitioners should align protection strength with data sensitivity, not interface convenience.

What this signals

Coverage depth is becoming a measurable security control. As agentic AI takes over repetitive inspection, the real question for testing programmes is whether neglected code surfaces still get examined before an attacker reaches them. That means pentest quality will increasingly depend on how well teams combine machine-assisted breadth with human judgment.

Whistleblower portals need stronger access governance than ordinary content areas. When a sensitive reporting path can be reached through a short code and exposed logic, the control model is too brittle for the data it protects. Security and identity teams should treat those flows as high-trust workflows, not convenience features.

For practitioners

• Prioritise large client-side bundles for AI-assisted review Run agentic analysis across oversized JavaScript and other low-priority code surfaces early in the engagement so obscure workflows are not left for spare time at the end.
• Test report-code protections for brute-force resistance Validate whether short access codes have rate limiting, lockout logic, telemetry, and alerting strong enough to stop automated guessing against confidential portals.
• Treat sensitive portals as access-governed assets Review whistleblower and other confidential submission flows as entitlement problems, not just UX features, and map every recovery or lookup path to explicit authorization.
• Reserve human effort for validation and impact Use the AI to surface candidate flaws, then spend analyst time proving exploitability, chaining paths, and documenting real business impact for the client.

Key takeaways

• Agentic AI helped surface a whistleblower report exposure path that human-only timeboxing would likely have left untested.
• The exposed portal relied on a weak six-digit alphanumeric protection scheme that could be brute-forced.
• Security teams should treat large client-side bundles and sensitive portals as priority targets for AI-assisted review and stronger access control.

Key terms

• Agentic AI: Autonomous AI systems capable of planning, deciding, and taking actions, including calling APIs, writing code, and orchestrating other agents, with minimal human oversight. Agentic AI introduces new NHI risks as agents must authenticate to external services.
• Timeboxed Pentest: A penetration test constrained by a fixed engagement window, often forcing prioritisation between breadth and depth. The security risk is not just missed findings, but the systematic exclusion of low-visibility code paths that can still contain high-impact flaws.
• Authorization: Authorization is the decision about what an authenticated identity is allowed to do. In NHI and IAM practice, it covers scope, duration, and allowable actions, and it is the layer that most directly controls blast radius when access is active.
• Client-Side JavaScript: Client-side JavaScript is code that runs in the user’s browser or mobile application rather than only on a protected server. Because it is delivered to the endpoint, attackers can inspect, modify, and reuse it. For security teams, that makes client-side logic a control boundary that needs protection, not just a development convenience.

Deepen your knowledge

The NHI Foundation Level course, the industry's only accredited NHI security programme, covers NHI governance, machine identity security, and secrets management. It helps practitioners connect identity control to the broader security work their programmes depend on.