TL;DR: AI security tools can spot problems but still fail to enforce governance across the full AI surface, Singul AI reports. Agentic workflows need execution-time control and traceable scope, or approvals and boundary checks remain inconsistent.
Editorial analysis by NHI Mgmt Group, based on content published by Singul AI: “AI Security Software vs a Unified AI Control Plane: The Fragmented Ecosystem Problem”.
Key questions
Q: Why do AI agents create governance risk even when they are meant to help testing?
A: AI agents create risk because they can follow unexpected paths, misread context, or act on malicious instructions without needing human intent.
Q: How do security teams know whether AI traffic controls are actually working?
A: They should look for evidence that policy decisions are consistent across prompts, responses, and tool calls, and that every block, mask, or alert can be traced back to a specific interaction.
Q: What breaks when data protection is split across SaaS, endpoint, browser, and AI tools?
A: Control drift breaks first.
Practitioner guidance
- Define the full AI surface before consolidating tools Inventory sanctioned SaaS AI, embedded AI features, browser assistants, endpoint agents, in-house applications and MCP-connected tools so governance decisions are made against the complete workflow path.
- Move policy checks to execution time Require every AI action that crosses systems to pass an explicit policy decision before the request is allowed to proceed, rather than relying on downstream alerts.
- Bind agent authority to task scope Assign each agent a narrow identity, limit it to the specific task it is performing and retire or revoke that authority as soon as the task completes.
Bottom line: The article’s core claim is that AI security fails when governance is split across disconnected tools and no single control can enforce the boundary of a workflow.
Glossary terms: Atlas Control Plane, Agentic workflow, Governance intent, Runtime Governance
Explore further
View Full Forum → | NHI Foundation Course → | Our Services → | Read the full analysis →
Fragmented AI security is a control problem, not a coverage problem. A stack that adds AI features to unrelated products may increase detector count, but it does not create a shared governance model. The issue is not how many alerts the organisation can collect, but whether any control can enforce the same policy across agents, endpoints, SaaS and cloud paths. Practitioners should judge AI security by enforcement scope, not by product adjacency.
A few things that frame the scale:
- 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.
- 19% of organisations give AI systems dramatically more access than human employees, nearly one in five granting unrestricted privilege, according to the 2026 Infrastructure Identity Survey.
A question worth separating out:
Q: What should teams do when an AI agent’s delegation chain crosses multiple systems?
A: Treat the full chain as one governance decision rather than separate local approvals. Each hop should be traceable to the same sponsor, purpose, and approved scope, with an escalation path when the combination would not be acceptable if requested directly. That is how separation of duties survives agentic execution.
👉 Read our full editorial: Fragmented AI security tools leave governance gaps across agentic workflows