Zero-Trust Agents for Autonomous Networks addresses one of the biggest barriers to scaling AI agents in telecom: how to trust autonomous systems in live operations. As CSPs move toward Level 4 Autonomous Networks, the question is no longer whether AI agents can automate decisions. The real question is whether operators can verify who is acting, control what is allowed, observe what happens across the full interaction chain, and retain evidence of every important outcome. This Catalyst responds to that challenge with a reusable, vendor-neutral zero-trust pattern for governed production. The project applies AI Governance via an IPOE model — Identity, Policy, Observability and Evidence — to agentic telecom operations. In practice, this means only approved agents are allowed into production, every action is bounded by delegated authority and policy, runtime behaviour is visible across agent, tool and model interactions, and every critical decision is backed by end-to-end evidence. The Catalyst demonstrates this approach in realistic autonomous network use cases, including governed autonomy, impostor agents, prompt injection, lateral movement and scope creep. It shows how operators can move from isolated AI experimentation to governed production with approved agents, delegated authority and provable outcomes. This is not just a technical issue. It is also a Layer 8 challenge of trust between people, systems and organisations. By combining runtime control with end-to-end accountability, the Catalyst provides a practical reference pattern that CSPs, vendors and partners can adopt to reduce risk, improve auditability and scale autonomous operations with greater confidence. It also feeds real-world implementation insight back into TM Forum standards development.