Skip to content
ModulesModule 08

Agent Trust Runtime

Enforce the policy rather than document it: sit in the path of the action and allow, restrict, escalate, deny or quarantine it.

What is it doing right now, and should this action be allowed?Phase 2 · Runtime governance

What it does

Capabilities

  • A gateway between the agent and the tool, API or database it is calling
  • A recorded decision per action, carrying agent, user, resource, context, risk, policy version and outcome
  • Behavioural baselines, so a hundredfold jump in record reads reads as an anomaly rather than a busy afternoon
  • Runaway protection that suspends an agent stuck in a loop before it exhausts cost or capacity
  • Human approval for sensitive actions, and two-person approval for critical ones

What it leaves behind

Evidence produced

  • An allow, deny or escalate decision on every governed action
  • A suspension when behaviour leaves the baseline

What the detection engine watches

Each of these is a departure from a baseline the platform established itself, not a threshold somebody guessed at during onboarding.

Privilege anomalyUnusual data accessUnusual volumeNew destinationNew MCP serverNew toolAgent loopCredential misuseModel changePermission driftOwner missingExpired certification

Specification

Where this is specified

5 blueprint pages carry the specification for this module.

21ENFORCEMENT

Runtime Agent Gateway

The runtime gateway turns Agent Trust Cloud from a dashboard into control infrastructure. It intercepts or brokers sensitive agent actions, enriches them with identity and risk context, evaluates policy, optionally requests approval, then forwards or blocks the action.

ENFORCEMENT PATH

22ENFORCEMENT

Agent Firewall

The Agent Firewall is the policy enforcement product experience layered on the gateway. It should make autonomous actions governable using understandable rules, not force customers to reason about low-level proxy configuration.

PRODUCT MOMENT

23AI SECURITY

Prompt-Injection Defense

Prompt injection is dangerous when untrusted content can influence an agent that has tools or authority. The defense should not rely on a single classifier. Use layered controls that reduce the consequences of manipulated instructions even when detection is imperfect.

OWASP ALIGNMENT

27AI SECURITY

Sandboxing & Execution Boundaries

Some agent tasks require executing code, browsing the web or handling untrusted files. These workloads need isolation because policy alone cannot guarantee that arbitrary code or content behaves safely.

ISOLATION PRINCIPLE

37SECURITY ANALYTICS

Behavioral Anomaly Detection

Behavioral detection identifies suspicious deviations that static policy cannot anticipate. The design should combine understandable statistical baselines with optional machine-learning models and never let anomaly scoring alone execute irreversible remediation without policy.

DETECTION RULE

13 modules and services make up the platform. See how they fit together.