Machine-speed adversaries are widening the pentest coverage gap

At a glance

What this is: Synack argues that machine-speed adversaries are exposing a pentest coverage gap, with only 32% of enterprise attack surface tested each year on average.

Why it matters: For IAM and security teams, the issue is that faster adversaries and AI-assisted testing change how risk is discovered, prioritised and validated across identity, cloud and application surfaces.

By the numbers:

• On average, only 32% of enterprise attack surfaces get tested in a given year.
• 64% of security leaders prefer agent-led testing with human oversight.
• 55% of security leaders say traditional testing fails to communicate findings in a way their teams can act on.

Context

Penetration testing is meant to validate whether real-world attack paths still exist, but annual or quarterly cycles rarely keep pace with fast-changing infrastructure and identity sprawl. When the attack surface expands faster than testing coverage, validation becomes partial by default, and the gaps tend to persist in the systems that matter most for access, privilege and lateral movement.

This article is about agentic AI changing the economics of offensive testing, not about replacing human judgment. The governance problem is that security leaders now have to distinguish real adaptive testing from automated scanning and polished reporting, while also deciding how to turn a larger volume of findings into prioritised action. That intersects directly with identity governance where credentials, privileges and trust relationships sit inside modern attack chains.

For identity and security programmes, the underlying challenge is not only discovery but decision quality. More testing can reveal more issues, yet without context the organisation may simply create more noise, more backlog and more unresolved risk than before.

Key questions

Q: Where does traditional pentesting fail in practice?

A: It fails when validation is periodic while the attack surface is continuous. Coverage becomes a sampling exercise, so exploitable paths can persist between test cycles. The practical failure is not only missed vulnerabilities, but missed chains across identity, cloud and application layers that a narrow engagement never exercises.

Q: Why does more testing create a prioritisation problem?

A: Because discovery scales faster than human triage. Once coverage improves, teams uncover more issues than they can remediate if they rely on raw severity scores alone. Context such as exploit likelihood, known exploitation and asset criticality becomes essential to separate noise from actionable risk.

Q: What do teams get wrong about testing agentic AI behaviour?

A: Teams often test the response and ignore the sequence that produced it. That misses retrieval failures, tool misuse, and unsafe intermediate steps that can still end in a plausible result. Effective testing must score the trajectory, session context, and policy boundary as a single control problem.

Q: How should organisations use human oversight in AI-assisted pentesting?

A: Human oversight should remain the decision layer for validating attack chains, business logic and whether a finding is truly exploitable. AI can expand reach and speed, but humans still need to judge context, confirm impact and prevent teams from acting on noise instead of risk.

Technical breakdown

Why annual pentesting cycles create coverage blind spots

Annual or quarterly pentesting assumes the attack surface is stable enough that periodic validation is meaningful. In practice, cloud assets, SaaS integrations, identity trust paths and exposed services change continuously, so a test report can be outdated almost as soon as it is delivered. Agentic testing changes the economics by increasing discovery speed, but it does not remove the governance problem of deciding what should be tested, what counts as coverage, and which findings are truly exploitable. The control issue is not just test frequency, but whether the programme can continuously map live assets to meaningful validation scope.

Practical implication: Treat test coverage as a living control metric, not a yearly project deliverable.

Agentic AI, scanners and generative AI are not the same control model

The article draws a useful distinction between automation, generative AI and agentic AI. Automation searches for known patterns at speed, generative AI mainly improves reporting and presentation, and agentic AI can plan, pivot and change strategy when the first approach fails. That distinction matters because only the last category can emulate adaptive adversarial reasoning across multiple systems and trust boundaries. For security governance, the question is whether a tool changes the quality of validation or merely accelerates output from the same underlying logic.

Practical implication: Evaluate offensive tooling by adaptive behaviour, not by interface polish or report volume.

Prioritisation becomes the bottleneck once testing expands

As testing reaches more of the environment, the challenge shifts from discovery to triage. The article notes that CVSS alone is not enough, which reflects a broader issue in security operations: a long list of findings is not the same as a usable risk queue. Context signals such as exploit likelihood, known exploitation and asset criticality help convert findings into action. That is especially relevant where identity and access controls are involved, because a low-severity flaw on a privileged path can matter more than a high-score issue on an isolated system.

Practical implication: Use exploitation context and asset criticality to sort findings before they hit remediation queues.

Threat narrative

Attacker objective: The attacker objective is to reach exploitable pathways that defenders have not validated, then turn those paths into real access before the next testing cycle catches up.

1. Entry begins with adversaries using AI-assisted discovery to identify exposed services and weak paths across a much larger attack surface than traditional teams can test regularly.
2. Escalation follows when chained vulnerabilities, trust relationships or privilege gaps are combined into a working path that scanners alone may not recognise as dangerous.
3. Impact is the persistence of untested, exploitable paths inside production environments, which gives attackers a faster route to compromise than periodic validation can close.

Breaches seen in the wild

• Meta Muse agent hijack 2026: An undocumented Muse setting let local malware hijack Meta's personal AI agent, steal its authentication material and abuse user access.

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

Coverage gap, not raw vulnerability volume, is the real control failure: annual pentesting leaves too much of the live estate outside validation for too long. When only a third of the surface is tested, the organisation is not measuring resilience so much as sampling it. The practitioner conclusion is to treat coverage as an operational control tied to live asset change, not as a periodic assurance exercise.

Agentic AI changes the offensive tempo, but not the need for human judgment: adaptive testing can chain findings and pivot when the first path fails, which is exactly where conventional scanners stop. That makes the human role more, not less, important, because someone still has to decide whether the chain represents real exploitable risk. The practitioner conclusion is to combine machine-scale discovery with human validation of attack paths.

Identity relationships become more exposed when testing becomes more continuous: more aggressive validation surfaces privilege misalignment, over-trusted service paths and weak handoffs between systems. In practice, this is where machine-speed adversaries and NHI governance intersect, because access paths are often the shortest route from discovery to impact. The practitioner conclusion is to govern credentials, authorisation and trust chains as part of the testing programme.

Detection quality matters as much as offensive capability: if security teams cannot separate automated output from genuinely agentic behaviour, they will misread both risk and control maturity. That creates decision debt across the programme, because leaders may think they are closing gaps when they are only producing more findings. The practitioner conclusion is to measure outcomes by confirmed risk reduction, not tool excitement.

From our research library:

• 67% of organisations still rely heavily on static credentials despite the risks they pose to agentic AI deployments, according to the 2026 Infrastructure Identity Survey.
• 53% of security leaders expect AI to run major portions of their infrastructure autonomously within the next three years, according to the 2026 Infrastructure Identity Survey.
• Read next: Agentic AI Security Guide

What this signals

Machine-speed validation will force security programmes to move from scheduled testing to continuous assurance: once offensive tooling can pivot faster than a quarterly cycle, the programme has to know where exposed identity paths and critical services change in near real time. That shift matters because the oldest pentest assumption is that a report remains useful for long enough to guide the next control decision.

Coverage is becoming an identity governance issue as much as a testing issue: when service accounts, delegated access and trust relationships are part of the attack path, the strongest validation programmes will be the ones that can show which privileges were actually exercised. The control boundary moves from finding flaws to proving that identity paths cannot be abused without detection.

The emerging pattern is a coverage trust gap: organisations may believe they are testing broadly enough, but they often cannot tell whether AI-assisted findings represent real exploitability or just faster output. That means the next generation of assurance needs better evidence quality, not just higher testing volume.

For practitioners

• Define live attack-surface coverage targets Set coverage metrics against current asset inventories, cloud accounts, exposed services and critical identity paths so testing tracks change rather than calendar cadence.
• Separate agentic testing from scanner output Require vendors and internal teams to prove that a system can pivot when the first approach fails, then document where human validation remains mandatory.
• Prioritise findings with exploitability context Rank discovered issues using EPSS, KEV status and asset criticality before they enter remediation queues, especially when identity or privileged paths are involved.
• Track identity and privilege paths inside testing scope Include service accounts, delegated access and trust relationships in recurring validation so the programme sees where lateral movement could start.
• Measure whether reports change decisions Test whether findings are producing contained remediation, reduced dwell time and fewer repeat exposures, not just higher report volume.

Key takeaways

• Annual pentesting cycles leave too much of the live attack surface outside validation, so the main risk is not just missed vulnerabilities but missed chains across systems.
• Agentic AI matters because it can adapt when the first attack path fails, which is a materially different control model from scanning or report generation.
• Security leaders need to prioritise findings using exploitability and asset context, or more testing will simply create a larger backlog of unresolved risk.

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.
• Attack Surface Coverage: Attack surface coverage is the share of a target system's reachable components that a test meaningfully examines. It is not just enumeration of assets. It reflects whether the testing process actually reaches the endpoints, workflows, and identities most likely to contain exploitable weakness.
• Exploitability context: Exploitability context is the evidence used to decide whether a vulnerability matters in a specific environment. It includes reachability, code path exposure, compensating controls, and product-specific advisories, and it turns raw scan data into a decision that can be defended.
• Adaptive Testing: A testing approach that adjusts execution, prioritisation, or maintenance based on application change and operational context. It is designed for environments where static scripts break quickly, and where validation must stay aligned with release velocity, infrastructure variability, and production-like conditions.

Deepen your knowledge

NHI Foundation Level course, the industry's only accredited NHI security programme, covers NHI governance, agentic AI identity and machine identity security for practitioners building stronger access controls. It helps security leaders connect identity governance to the broader programmes they run across cloud, application and infrastructure.