Scenario-based demonstrations

Demonstrations

• Supplier payment approval

  About 10 minutes

  Architecture

  Finance agent → Python agent SDK → CydraGateway → finance tool (simulated).

  Data flow

  The agent requests a £48,500 supplier payment with its workload token and an idempotency key.

  Policy decision

  Policy requires approval above the payment threshold; a named approver approves with a single-use token.

  Outcome

  The payment executes exactly once, and the decision maps to EU AI Act Article 14 human-oversight evidence.

• Containing a compromised agent

  About 5 minutes

  Architecture

  CydraShield kill switch acting on CydraGateway identities and credentials.

  Data flow

  An agent with a high risk score is suspended from the agent page.

  Policy decision

  Pending and approved actions from the contained agent are revoked before execution.

  Outcome

  Further requests from the agent are denied, and the containment is recorded as evidence.

• Mapping an agent's blast radius

  About 5 minutes

  Architecture

  Connectors → agent inventory → Agent Security Graph.

  Data flow

  Discovered identities, delegations, MCP servers and tools are projected into the graph.

  Policy decision

  The Agent Risk Score explains which factors drive the agent's risk and which controls reduce it.

  Outcome

  You see every data source the agent can reach, directly or transitively, and the highest-risk path.

• Evidence for an audit

  About 5 minutes

  Architecture

  CydraGateway evidence chain → CydraGovern control mapping → report export.

  Data flow

  Decision records are linked to controls through mapping rules.

  Policy decision

  The evidence chain is verified end to end before the report is produced.

  Outcome

  An audit-readiness view of control coverage, evidence freshness and open gaps, exportable as a report.

Demonstrations use sample data, not customer data. Recorded versions will be added here.