The organisation remains accountable because the failure is governance, not just tooling. Personal accounts and unmanaged keys remove revocation, attribution, and auditability, which weakens both security and compliance. Platform engineering, security, and identity teams should own the control design, while line teams must follow the approved enterprise identity and secrets process.
Why This Matters for Security Teams
When Claude Code is accessed through personal accounts or unmanaged API keys, the risk is not limited to one developer’s workspace. It creates a control gap around identity, revocation, logging, and ownership, which means the organisation loses the ability to reliably answer who did what, when, and under which authority. That is why this sits squarely in governance and access management, not just developer convenience.
Security teams should treat the issue as an enterprise control failure that can affect software supply chain integrity, data handling, and incident response. The NIST Cybersecurity Framework 2.0 is useful here because it frames accountable operations around governance, identity, protection, and recovery outcomes. For AI-enabled tooling, the lesson is sharper still: identity sprawl and unmanaged secrets can undermine both tool access and downstream actions taken by an agent or developer workflow. In practice, many security teams encounter the accountability problem only after a key has been copied into a personal environment and used long enough to blur ownership, rather than through intentional enterprise access design.
How It Works in Practice
Accountability depends on whether the organisation can bind access to a managed identity, enforce policy, and preserve audit evidence. If a user authenticates with a personal account, the enterprise may lose centralised lifecycle control even if the work is done for business purposes. If an API key is issued outside the approved secrets process, the key may persist beyond the role, project, or employment relationship that justified it.
Good practice is to ensure Claude Code access is provisioned through enterprise identity, with role-based approval, least privilege, and revocation workflows. Secrets should be stored and rotated through approved systems, not copied into local files, browser profiles, chat tools, or ad hoc scripts. Logs should record who requested access, who approved it, which system used the credential, and what scope was granted. The OWASP Non-Human Identity Top 10 is directly relevant because unmanaged API keys behave like unmanaged non-human identities: they can outlive their owners and evade normal joiner-mover-leaver controls.
- Bind tool access to corporate identity, not personal email or consumer sign-in.
- Issue API keys through a central secrets platform with expiration and rotation.
- Use separate credentials per environment, workload, or team.
- Monitor for shadow usage, copying of keys, and long-lived tokens.
- Require revocation playbooks for offboarding, compromise, and project end.
For AI-assisted development and agentic workflows, the control objective should also include action traceability and tool-scoped permissions, because a compromised key can authorise both code access and automated follow-on actions. These controls tend to break down in fast-moving developer environments with local experimentation, shared terminals, and informal key exchange because those conditions bypass the enterprise approval and logging path.
Common Variations and Edge Cases
Tighter identity and secrets control often increases friction for developers, requiring organisations to balance speed against auditability and revocation. That tradeoff is real, and best practice is evolving in some areas, especially where teams are adopting AI coding assistants, ephemeral environments, and agentic workflows that need temporary access.
One common edge case is contractor access. A contractor may need rapid onboarding, but accountability still requires corporate sponsorship, time-bound access, and explicit ownership of the credential path. Another is break-glass access for incidents. Current guidance suggests that emergency access can be justified, but it must be tightly logged, time-limited, and reviewed after use. A third edge case is delegated automation, where a team believes a tool is “owned” by the vendor or platform. That is not a valid accountability transfer: the enterprise remains responsible for the decision to permit the access and for the controls around it.
Anthropic’s report on its first AI-orchestrated cyber espionage campaign shows why tool access and identity governance matter once AI systems can be used operationally for actions, not just suggestions. The operational takeaway is to treat unmanaged accounts and keys as a control exception requiring explicit risk acceptance, not as a harmless productivity shortcut. Where teams rely on local developer machines, multi-account cloud setups, or overlapping personal and corporate sign-ins, the accountability model becomes ambiguous and the control set loses enforceability.
Standards & Framework Alignment
This section maps relevant standards and security frameworks to the operational risks and controls described in this guidance.
OWASP Non-Human Identity Top 10 address the attack and risk surface, while NIST CSF 2.0, NIST AI RMF and NIST SP 800-53 Rev 5 set the governance and control requirements practitioners need to meet.
| Framework | Control / Reference | Relevance |
|---|---|---|
| NIST CSF 2.0 | GV.OC, PR.AC | Governance and access control frame ownership, approval, and revocation for this access issue. |
| OWASP Non-Human Identity Top 10 | NHI-01 | Unmanaged API keys act like non-human identities without lifecycle control or attribution. |
| NIST AI RMF | GOVERN | AI governance must define responsibility for tool access and actions taken by AI-enabled workflows. |
| NIST SP 800-53 Rev 5 | AC-2 | Account management controls are needed when access is tied to personal or unmanaged identities. |
Use managed accounts, approval workflows, and prompt deprovisioning for every user and service identity.
