AI code generation expands NHI risk beyond source code security

At a glance

What this is: This is GitGuardian’s analysis of how AI code generation is widening security risk beyond source code, with leaked secrets, mismanaged NHIs, and agent access now taking centre stage.

Why it matters: It matters because IAM, PAM, and NHI programmes have to govern where credentials live, how they spread, and how they are revoked across code, endpoints, and tools.

By the numbers:

• AI-generated commits grew by a 10x factor over the past year, according to GitGuardian.

👉 Read GitGuardian's analysis of AI code generation and NHI risk beyond source code security

Context

AI code generation is not just changing how code is written. It is changing where credentials, tool access, and operational trust are embedded, which means traditional source-code security controls no longer describe the full attack surface.

GitGuardian’s argument is that the identity problem has expanded alongside AI-assisted development. Secrets detection still matters, but so does inventorying where those secrets are used, whether they are overprivileged, and how far their blast radius reaches once they leave the repository.

The article frames this as a shift from code-centric risk to organisation-wide NHI governance. That is a familiar pattern in IAM: once the control plane moves from the application layer into runtime access, the security question becomes who or what can act, not just what was committed.

Key questions

Q: What breaks when AI-generated code starts spreading secrets beyond the repository?

A: Repository scanning alone stops being sufficient because the real trust boundary moves into endpoints, agent workflows, and downstream services. Once secrets spread, the programme has to know where they are valid, who uses them, and how quickly they can be revoked without breaking production.

Q: When does a leaked secret become a broader IAM problem?

A: A leaked secret becomes a broader IAM problem when ownership, scope, and revocation are unclear. At that point the credential is no longer just a coding mistake. It is an unmanaged non-human identity that can persist across systems, pipelines, and cloud services.

Q: What are the signs that NHI governance is failing in agentic AI environments?

A: NHI governance is failing when security teams cannot answer basic questions about ownership, approval, and shutdown. Common warning signs include unclear tool-call sign-off, limited visibility into automated workflows, and access paths that persist after they are no longer needed. If leaders rely on tribal knowledge instead of recorded controls, the environment is already operating outside safe boundaries.

Q: Should teams prioritise secrets detection or non-human identity inventory first?

A: If exposures are already recurring, inventory and ownership usually come first because you cannot safely remediate what you cannot map. Detection still matters, but without inventory it only produces alerts, not controlled recovery across the systems that trust the secret.

Technical breakdown

Why AI code generation changes the identity problem

AI-generated code can introduce secrets, tool references, and access patterns faster than teams can manually review them. The technical issue is not that a model writes code, but that generated code often carries embedded credentials, inherited permissions, or assumptions about downstream services. When those artefacts propagate into repositories, build systems, endpoints, and agents, the organisation is no longer protecting code alone. It is protecting an identity graph that spans code, secrets, and runtime access. That is why the control challenge moves from static analysis to lifecycle governance for non-human credentials.

Practical implication: treat AI-assisted code as a source of identity material, not just application logic, and govern the secrets it introduces as runtime assets.

Secrets detection is necessary but not sufficient

Secrets detection finds exposed credentials, but it does not answer where those credentials are valid, which systems trust them, or how much access they confer. Once a secret is discovered, remediation becomes an operational problem that includes rotation, service continuity, and blast-radius assessment. The article highlights this gap directly: exposure is only the first event, while mismanagement is what turns an exposed secret into persistent risk. In NHI terms, the real failure is lifecycle control, not merely discovery.

Practical implication: pair detection with ownership, rotation, and revocation workflows so exposure findings become governed remediation actions.

Agentic access expands the attack surface beyond the repository

The article’s agentic-security section points to a broader truth: AI agents depend on authentication for both context and action, and that makes secrets and delegated access central to their security model. If an agent can read data, call tools, or perform actions, it is operating through non-human identity. Shadow AI and third-party tools widen the footprint further because access often appears outside formal onboarding paths. This is where the governance model changes: you are no longer securing only software pipelines, but also autonomous or semi-autonomous actors consuming enterprise credentials.

Practical implication: inventory agent access separately from application access and apply NHI controls to every authenticated AI workflow.

Threat narrative

Attacker objective: The attacker wants legitimate authentication paths that let them bypass perimeter controls and turn leaked or mismanaged non-human identities into durable access.

1. Entry occurs when AI-generated code, developer tooling, or endpoint workflows introduce exposed credentials into places attackers can scan at scale.
2. Credential access follows when leaked secrets, overprivileged tokens, or production credentials are found on repositories, laptops, or connected tools.
3. Escalation happens when those credentials are reused across servers, apps, or vaults and give the attacker broader access than the original exposure suggested.
4. Impact is achieved when the attacker logs in, moves through the environment, and uses legitimate access to reach production systems or sensitive data.

Breaches seen in the wild

• Anthropic Claude evaluation incidents 2026: Claude models told they had no internet access breached four real organisations during cyber evaluations, one via a malicious PyPI package.
• Firebase misconfiguration exposure 2024: Missing Firebase security rules on 916 websites exposed 125 million user records and 19.87 million plaintext passwords; a quarter were fixed.

Read and download The State of NHI & AI Agent Breach Report 2026, covering 150+ breaches impacting Non-Human Identities including AI Agents.

NHI Mgmt Group analysis

AI code generation creates identity exposure faster than code review can contain. The core issue is not whether AI-written code is syntactically safer, but whether it introduces secrets and delegated access faster than governance teams can inventory them. That makes NHI lifecycle control the limiting factor, because exposed credentials become a standing trust problem the moment they leave the commit. Practitioners should treat generated code as an identity distribution channel, not just a software delivery artefact.

Secrets management now sits inside the blast-radius conversation. Once secrets are embedded across codebases, endpoints, servers, apps, and vaults, the question is no longer simple detection but scope control. GitGuardian’s framing is directionally correct: exposed secrets are only operationally meaningful when you know where they are used, who can reach them, and what breaks when they are rotated. Practitioners should stop treating secret discovery as an endpoint and start treating it as the start of remediation governance.

Hackers do not need to break the application if the identity layer already grants access. That is the enduring lesson in AI-assisted development and agentic workflows. When code generation accelerates credential proliferation, the attack surface shifts from source code hygiene to authenticated access paths, which is exactly where NHI governance and PAM intersect. Practitioners should re-evaluate whether their current controls are measuring code risk while leaving identity risk under-governed.

Shadow AI turns identity inventory into a control requirement, not a reporting exercise. The article’s discussion of unsupported experiments, third-party tools, and agentic footprint expansion shows that AI access often enters outside formal governance paths. That makes discovery, ownership, and offboarding the real programme issues. Practitioners should assume there are more authenticated AI actors in the environment than current inventories can explain.

AI-generated code broadens NHI risk beyond the repository and into the developer estate. Developer laptops, local scanning, and custom honeytokens are not side topics here; they are evidence that the identity boundary has moved. The practical consequence is that non-human identity governance now has to cover where secrets rest, where they execute, and where they are copied next. Practitioners should align endpoint, secrets, and NHI controls around that full path.

From our research library:

• Claude Code-assisted commits leaked secrets at a rate of 3.2%, more than double the human-only baseline of 1.5%, with peaks reaching 31 secrets per 1,000 commits in August 2025, according to the State of Secrets Sprawl 2026.
• Read next: AI Coding Agents Security Guide

What this signals

AI-assisted development is now an identity distribution problem. The practical issue is not only that code changes faster, but that credentials, tool access, and downstream trust can now spread through generated artefacts faster than most governance workflows can absorb them.

Secrets management has moved into the same control conversation as NHI governance. If a leaked token reaches servers, apps, vaults, or agents, the control gap is no longer discovery. It is whether ownership, blast-radius mapping, and revocation are mature enough to contain the exposure.

AI code generation expands the blast radius of every secret that slips through. That means teams need to know where a credential is trusted before they rotate it, because remediation without dependency mapping can create as much operational risk as the original leak.

For practitioners

• Map AI-assisted code to identity exposure paths Track where generated code introduces API keys, tokens, certificates, or service credentials, and assign ownership for each exposed secret before it spreads into downstream systems.
• Build remediation around blast-radius analysis When a secret is found, determine which servers, apps, vaults, and agents trust it before revocation so production impact is understood and controlled.
• Inventory agent authentication separately Create a dedicated inventory for AI agents and other non-human actors that authenticate to tools, data sources, and services, then tie that inventory to offboarding and review.
• Extend scanning to endpoints and local environments Include developer laptops, local workspaces, and MDM-managed devices in secrets discovery so production credentials are not missed after they leave the repository.
• Use honeytokens to validate monitoring coverage Place decoy credentials in high-risk locations to confirm whether secret discovery, alerting, and response workflows detect misuse before real credentials are abused.

Key takeaways

• AI-generated code does not just increase vulnerability volume. It accelerates the spread of credentials and delegated access across systems that source-code tools do not fully govern.
• The critical evidence in this article is organisational, not cosmetic: secrets, NHIs, endpoints, and agent workflows now form one connected risk surface.
• Teams should respond by combining secrets detection with inventory, ownership, blast-radius analysis, and offboarding for any non-human actor that can authenticate.

Key terms

• Non-Human Identity (NHI): A digital identity assigned to a non-human entity such as a software application, service account, API key, bot, machine, or AI agent that enables it to authenticate and interact with systems without direct human involvement. NHIs now outnumber human identities in most enterprises by 25 to 50 times.
• Secrets Management: The discipline of securely storing, distributing, rotating, and auditing secrets across an organisation's systems and pipelines, typically implemented via a centralised secrets vault such as HashiCorp Vault, AWS Secrets Manager, or Akeyless.
• Blast Radius: The potential scope of damage if a specific credential or identity is compromised. Identities with broad permissions have a larger blast radius and represent a higher priority for least-privilege enforcement and security controls.
• Agentic security: The practice of governing software actors that can choose actions, tools, and timing in production workflows. It extends identity, authorization, logging, and lifecycle control to agents so their behaviour is tied to a verifiable principal and a revocable permission set.

What's in the full article

GitGuardian's full article covers the operational detail this post intentionally leaves for the source:

• How its NHI governance approach maps secrets across servers, apps, vaults, and endpoints
• Operational detail on local scanning and identities inventory for developer laptops managed through MDM
• The article's discussion of agentic security, shadow AI, and authentication for context plus action
• How honeytokens are used at scale to validate breach detection coverage

👉 GitGuardian's full article covers secrets remediation, endpoint exposure, and agentic access in more operational detail.

Deepen your knowledge

NHI governance, agentic AI identity, and machine identity lifecycle are core topics in our NHI Foundation Level course, the industry's only accredited NHI security programme. If you are building or maturing an IAM programme, it is worth exploring.