Developer Access Risks Grow as AI Agents Multiply
A BlueFlag Security report finds developer and AI agent accounts often keep unused permissions that can reach critical code repositories.
Updated on Nov. 6, 2026 in Software
BlueFlag Security released The State of Developer Risk 2026, a report on how developer and AI agent identities access code repositories. The study finds permissions frequently outlast the work they were granted for, creating exposure across software organizations.
Why it matters
As coding assistants and autonomous agents write, review and merge code, the identities they operate under can carry broad access that few organizations actively monitor. The findings point to a widening gap between how software is built and how access to it is governed.
In the pooled dataset, 99% of repository-access tokens had no expiration date and 47% had been unused for more than 90 days without being revoked. Unapproved AI agents were present in 83% of organizations measured, though the report does not say how many of those agents caused security incidents.
The players
BlueFlag Security
BlueFlag Security is a cybersecurity company that publishes research on identity and access risks in software development environments.
The details
The report draws on developer and service account activity across organizations, examining the permissions and tokens that coding tools and autonomous agents use. Coding assistants typically work under a developer's own identity and inherit its permissions, while autonomous agents act as separate identities that can write and merge code without a human approving each action. Among the findings, four in ten repositories had unprotected main branches, and code was merged despite failed checks in 56% of organizations.
Timeline
November 6, 2026: Help Net Security publishes coverage of the BlueFlag Security report.
Big Picture
The BlueFlag report's measurement that non-human identities outnumber human developers eight to one extends the identity-sprawl problem into the AI agent era. Its finding that most access tokens never expire shows that the long-lived credential model built for human developers is now carrying autonomous software.
Developers and teams using AI coding tools may find that their own permissions are broader than their current work requires. Reviewing unused tokens and enforcing expiration dates are practical steps organizations can take now.
The takeaway
Regularly auditing which accounts, tokens and AI agents can reach code is a concrete step teams can take. Expiring unused credentials and requiring approval for new agents reduces exposure without slowing legitimate development work.
Further reading
Explore more on Software developments and how code security practices are evolving.
Source note: This article includes information reported by Help Net Security.
