Controlling and enforcing access to APIs and network-accessible MCP capabilities, with authentication, policies, least-privilege tool exposure, scopes and runtime controls.
Tyk × Venedy · Wiesn Breakfast
AI Governance
in Action.
- 09:00-11:00
- WERK1 Munich
An AI agent is allowed to use a tool. But is it allowed to read this customer’s data with it, or approve this transaction? We show live where such boundaries are enforced and how to test whether they hold.
The breakfast took place on 2 October 2026. Thank you to everyone who joined us! We will announce the next event on LinkedIn.
The central question
What can an AI agent access, change and trigger?
AI agents are moving beyond generating answers toward taking action. They access data, call tools and trigger workflows, often through APIs and increasingly through protocols such as MCP.
Risks include tool misuse, identity and privilege abuse, goal hijacking and unsafe delegation.
- Who is acting? What are they allowed to call?
- On which data, under which delegated authority and in which business context?
- How is that decision enforced? How do we test that the assumption is actually true?
Governance guides each step
- DefineDefine permitted APIs, MCP tools and actions.
- EnforceEnforce identity, authorization and least privilege.
- ObserveObserve invocations and preserve evidence.
- ValidateValidate downstream API authorization.
- ImproveRetest critical security assumptions.
We’ll discuss how to put these steps into practice, including tests of dangerous combinations of otherwise legitimate capabilities and permissions.
Venedy
Test authorization & provide evidence
Actively testing the downstream APIs and business logic across identities to identify authorization flaws and produce reproducible evidence that intended boundaries hold or demonstrate where they fail.
What we’ll discuss
Putting governance into practice and testing the results
A token may be valid. The agent may have the expected scope. The gateway may route the request correctly. Yet an underlying API can still expose another customer’s object, allow an unauthorized function or violate a business rule.
A permitted tool call does not automatically mean the underlying business action is authorized. Active security testing helps uncover these gaps.
- AI Governance: Who sets the rules for AI agents, and who owns them in the organization.
- API & MCP Security: Which tools an agent may see and call, and through which path.
- Authorization & Agent Identity: Whose identity actually reaches the backend.
- Security Validation: How to check that boundaries hold, and what a test really proves.
- AI Agent Risks: What real incidents reveal about typical misconfigurations.
- After every change: How boundaries hold as new tools and releases arrive.
The morning
A practical conversation
A focused morning for security, platform, engineering and technology leaders considering how to bring agentic AI into production without losing control over what those agents can actually do.
- 30 min
Arrival & breakfast
Coffee, pretzels and first conversations.
- 30 minEnforce & verify
Where agent permissions are enforced, and whether the boundary holds
From the agent’s call down to the single record: identity, tool approvals and limits, shown live on a working agent. And how to test whether the application really enforces the boundary.
- 10 min
Coffee break
- 40 minDiscuss
Open discussion
How are agent systems built, governed and controlled in practice? Room for experience, open questions and exchange with us and peers from different industries.
Wrap-up
Thank youThank you
Thank you
for joining us.
The breakfast took place on 2 October 2026 at WERK1 in Munich. Thank you to everyone who joined us and shared their questions and experience with us.
Questions or feedback about the breakfast?
event@venedy.io
Join us next time
We will announce the next event on LinkedIn. Follow Venedy there so you don’t miss it.