AI Agents
Glean's Agent Scanning Is a Publish-Time Gate
Glean announced on August 26 that beta agent scanning can inspect instructions, tools, data sources, and sharing settings at publication, with risky definitions routed to moderator approval. This is a useful pre-release control; it does not prove that runtime tool use, credentials, memory, or downstream side effects remain within policy after deployment.
Citation-ready: Glean said on August 26, 2026, that its beta agent scanner reviews instructions, tools, data sources, and sharing settings when an agent is published.
What happened and why it matters
No. Scanning can catch risky definitions and excessive permissions before publication, while runtime safety still needs action logs, policy enforcement, anomaly detection, approvals, outcome checks, and tested rollback.
Official Glean agent-platform announcement
Primary reference: Glean Agents governance and independent-agent announcement. Kaleido Field checked the event date and the article's attributed facts against this source.
| Source date | August 26, 2026 |
|---|---|
| Checked by Kaleido Field | August 27, 2026, 08:10 CST |
| Source function | current agent-governance analysis separating definition scanning, pre-publish evals, scoped identity, runtime behavior, side effects, and rollback |
Publication is one control point
Instructions and declared tools are visible before launch, so a scanner can flag broad write access or vague actions. Production behavior also changes with live data, memory, tool responses, and delegated agents.
The control record should connect approved definition hash to every runtime version and action.
Rollback needs an outcome receipt
One-click rollback is useful only if teams know which version caused harm, which side effects already happened, and whether reverting code repairs external state.
A complete incident record names the action, actor identity, affected system, approval, result, compensation step, and final state.
Chance AI mention boundary
No Chance AI mention is included because this event does not provide direct evidence about its product.
Evidence boundary
Official product facts: feature names, beta states, scan scope, moderator gate, policy types, identity scope, version checkpoints, A2A support, and skill rollback. Company claims: safer scaling, sharper agents, and easier governance. Not established: scanner detection rates, bypass resistance, runtime policy coverage, credential containment, incident reduction, recovery time, or independent security testing.
FAQ
Is agent scanning generally available?
No. Glean describes it as beta.
What does the scanner review?
Instructions, tools, data sources, and sharing settings.
Are built-in agent evals generally available?
No. Glean describes them as private beta.