BlueFlag Security has made new controls for AI coding agents generally available, including tools running locally on developers’ machines. The practical change is that security teams can find and review agents using developer credentials, rather than waiting for those agents to announce themselves politely.
Watch Desk analysis
What happened
SiliconANGLE reports that BlueFlag’s developer-risk platform now tracks agents through their code commits and pull requests, linking them to developers and repositories. To detect local tools, it draws on data from endpoint security products such as CrowdStrike and SentinelOne, without installing another agent on each device.
The update can flag pull requests approved by an AI agent, code written end-to-end by one, and pipeline runs without human sign-off. Teams can require a person to review changes before merge or deployment. BlueFlag also added inventories of AI models and frameworks in applications, tracking of AI spending against accepted code, and two remediation tools: AccessGuard for identity problems and DriftGuard for repository and pipeline settings. Fixes can be applied manually or with one click, and automated changes can be rolled back.
Our top picks
- Find agents beyond the code repository
Endpoint-security data can help surface AI tools running on developers’ machines. - See where human review happened
Teams can flag agent-approved pull requests and require review before code merges or deploys. - Trace AI use in applications
An inventory records the model providers and agent frameworks used in software a company builds. - Tie spending to output
BlueFlag links token and dollar use to commits and accepted code, and highlights rising costs without new commits. - Apply and undo selected fixes
AccessGuard and DriftGuard target identity and configuration problems, with automated changes that can be rolled back.
Why it matters
Coding agents can inherit the permissions of the developer whose credentials they use. That makes it harder for organisations to know which software acted, what it could access and whether a person reviewed the result. BlueFlag’s release turns those questions into things security teams can inspect and, in some cases, control.
The article also describes BlueFlag research based on platform metadata from a subset of its enterprise customers. It says unapproved coding agents appeared in 83% of the organisations measured, and autonomous agents had merged code without a person in 22%. Those figures describe that sample, not every organisation using coding agents.
Our read
This is a concrete answer to the gap between “we let developers use AI” and “we know what those agents can do”. The strongest bits are visibility into local tools and explicit human-review controls. BlueFlag says the new features are generally available; as ever, a dashboard is only as useful as the decisions teams make from it. Security teams should look closely at what the platform can actually detect in their own environment before treating the inventory as complete.
What to watch
- Whether customers report finding agents outside the code repository.
- How often teams require human review before merges and deployments.
- Whether BlueFlag delivers its planned attack-path interventions early next year.
Discussion spark: Should organisations require human approval for every AI-written code change, or is clear visibility and risk-based review enough?
Sources and evidence
- Exclusive: BlueFlag Security expands AI agent governance to cover unapproved coding agents – SiliconANGLE (4 November 2026, 14:00 UTC)
- Descope launches Envoy to let consumer apps accept personal AI agents as users – SiliconANGLE (4 November 2026, 14:00 UTC)
- Google fine-tunes its Agentic Data Cloud to support highly scalable autonomous AI workers – SiliconANGLE (4 November 2026, 14:00 UTC)
Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.