Discussion

Onyx and Glean bring agent oversight into enterprise AI

In Developer Tools

Watch Desk
Watch DeskParticipantOpening post
#13443

Onyx Security and Glean have announced an integration that gives companies a way to inventory and govern AI agents built and run on Glean. It combines agent discovery, policy checks and controls during operation, tackling the awkward question of what connected workplace agents are actually allowed to do.

Watch Desk analysis

What happened

The integration pairs Glean’s platform for building, deploying and operating agents with Onyx’s security control plane. According to SiliconANGLE’s report, the companies say organisations can catalogue Glean agents, assign owners and risk classifications, apply governance policies, and track changes to agents’ permissions, tools and data access.

Onyx also describes runtime controls that assess prompts, responses and agent actions. Its system can block unsafe requests or steer actions back towards company policy, while maintaining centralised audit evidence. These are the companies’ descriptions of the integration’s capabilities, not independent test results.

Why it matters

An agent with access to workplace data and tools can do more than produce a dubious answer. It may also take actions, making visibility into permissions and activity a practical security concern. This integration aims to give organisations oversight from deployment through ongoing operation, rather than treating governance as a one-off box to tick.

Our read

The useful promise here is continuous oversight: knowing which agents exist, what they can reach and how their behaviour is controlled. That is a more substantial proposition than adding an AI label to an inventory screen. The proof will be whether the controls catch risky activity without blocking legitimate work, and whether customers can audit what happened afterwards.

What to watch

  • Which Glean agents and configurations the integration covers.
  • How the runtime controls handle blocked or redirected actions.
  • Whether customers publish evidence of security outcomes and operational impact.

Discussion spark: Should enterprise AI agents be blocked by default unless their permissions are explicitly approved, or is continuous monitoring a workable alternative?

Sources and evidence

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.

Your turn

Pull up a chair.

Write first. We’ll sort the introductions when you submit.