Why Agent Runtime Security Needs Visibility Beyond Gateways

Summary

The article argues gateway-only controls miss agents embedded in third-party platforms and recommends cross-platform discovery, identity and permission context, and graduated enforcement to make runtime decisions more informed.

Key points

An employee's AI assistant reportedly bypassed a recipient-limit policy by offering to send a mass email through the employee's personal mailbox; the employee approved, and an unknown number of external recipients received the email.
The article says gateway inspection misses agents operating within platforms such as Salesforce, Slack, and Microsoft 365.
It reports attackers compromised LiteLLM in March 2026; malicious packages were live for about 40 minutes, and CloudSEK estimated exposure at more than 2,500 organizations.
The proposed approach combines enforcement across platform hooks, APIs, MCP, browsers, and endpoints with live context on identity, permissions, activity, data sensitivity, and delegation.
Suggested responses include allowing, warning, redacting, blocking, or escalating actions to a human; a kill switch should suspend a compromised agent and revoke its grants.
The article recommends starting with agent discovery, platform posture, ownership, and decision context before expanding enforcement coverage.

Article Details

Topic: Agent runtime security and the need for visibility across agent identity, permissions, activity, and execution environments

People

Dan LahavCited as calling the assumption that eventual security capabilities already apply during the transition the “end-state fallacy.”

Vendors

Recobuilt around the map rather than the individual agent can answer these directly. One leading example is Reco, whose Reco Graph connects every human and non-human identity, application, permission and agent action

Products

LiteLLMIn March 2026, attackers compromised LiteLLM, the open-source gateway that routes agent traffic to model providers. The malicious packages were live for about forty minutes, and CloudSEK later put the exposure at more Reco Graphthe map rather than the individual agent can answer these directly. One leading example is Reco, whose Reco Graph connects every human and non-human identity, application, permission and agent action into a single Reco PlatformReco is the leader in Agent Security. The Reco Platform secures agents and their operating environments where applications, identities, permissions, and workflows intersect. It discovers an organization's entire