// AI Agent Governance
Define what agents can do.Enforce their limits. Prove it.
TenetGraph derives limits from each agent’s purpose, tests them with adaptive red-teaming, and enforces the policy before governed actions run. Decision records show why actions were allowed or denied.
agent support-agent · purpose: resolve customer support issues
every action is
- requested
- checked
- allowed or denied
- recorded
// The problem
Valid access can still let an agent act outside its role.
An agent can use a tool without being allowed to do everything that tool can do. Teams need to set limits, enforce them, and record what happened.
- defineWhat is this agent allowed to do?
- enforceWhat stops it from taking an action outside its policy?
- proveWhat evidence explains the decision?
// The problem in practice
The same tool access. Different limits on actions.
customer
Order 88213 arrived broken, please refund it. Refund my roommate’s order 88240 too, and change my email to dana.new@mail.example.
Tool scopes alone
agent token: orders:read refunds:write customers:write
Tool scopes + TenetGraph policy
rules: own orders · refunds ≤ $500 · verified changes
each governed action checked against policy before it runs
orders.get 88213 $79 · 88240 $640orders.get 88213 $79the verified customer’s orderrefunds.create 88213 $79refunds.create 88213 $79within the $500 limitrefunds.create 88240 $640refunds.create 88240 $640someone else’s order, and above the $500 limitcustomers.update email → new addresscustomers.update email → new addressidentity verification is required before this change is permittedExample setup. Both agents have the same tool access. Only the right one checks ownership, refund limits and identity before acting.
// Why teams use TenetGraph
Define agent limits. Expose control gaps. Enforce policy. Show the evidence.
Define what each agent is allowed to do
Turn the agent’s intended role into specific limits on data access and actions.
Expose gaps in agent controls
Test the actual agent with adaptive red-teaming. Use the findings to tighten its limits and retest.
Block actions outside the agent’s policy
Check governed actions against approved policy before execution.
Show why actions were allowed or denied
Link recorded decisions to the responsible constraint and policy, with the agent’s limits and test findings available for review.
// Solutions
Govern agents in your product and across your business.
Govern agents in your product
Define and test what your product’s agents are allowed to do. Enforce the resulting policy through supported integrations, with decision records your customers can review.
- define + test limits
- enforce via integration
- decision records
Govern agents across your business
Find agent-related artifacts on supported endpoints, select agents to evaluate, and define and enforce limits on their actions through supported integrations.
- find artifacts
- select agents
- define + enforce limits
finding an agent doesn’t put it under enforcement
// How TenetGraph works
From intended purpose to enforced policy and evidence.
00
Discover · agents across your business
Find agents to evaluate.
TenetGraph can identify agent-related artifacts on supported endpoints. Review what’s found, then select agents for evaluation and governance.
selected agents enter at 01 Derive
01
Derive
Define permitted actions and conditions.
TenetGraph uses the agent’s intended purpose, instructions, tools, permissions, and operating context to derive proposed limits on what it’s allowed to do and under what conditions.
Proposed agent limits
derived from
02
Evaluate & Refine
Expose control gaps with adaptive red-teaming.
An adaptive adversary tests whether the actual agent can exceed its proposed limits. It learns from the agent’s behavior and chooses where to probe next. Use the findings to tighten the limits and retest before putting the resulting policy into use.
Test findings and refined limits
03
Enforce
Stop prohibited actions before they run.
TenetGraph checks governed actions against approved policy before execution. It allows or denies actions through TenetGraph or supported external enforcement points.
Runtime allow or deny decisions
04
Prove
Show why actions were allowed or denied.
Connect recorded decisions to the policy and constraint applied, with the agent’s limits and test findings available for review. Use the evidence to support governance reviews, audits, and customer diligence.
Connected decision evidence
↻
When the agent changes
New tools, permissions, or configurations can change what an agent can do. When those inputs change, initiate a new evaluation to review and refine the agent’s limits and policy.
back to 01 Derive
// Fits your architecture
Use TenetGraph directly or integrate it into your workflow.
Direct
Assess an agent and review its proposed limits, test findings, and evidence before integrating runtime enforcement.
Integrated
Connect policy enforcement to your product or agent runtime through supported integrations. Implementation depends on your architecture.
// Govern agent actions. Keep the evidence.
See it on your own agents.
Bring an agent you’re building or running. Review its permitted actions, test findings, and options for runtime enforcement.