Communications service providers need to know that autonomous AI agents are secure, compliant and aligned with business intent before they act. This Catalyst adds security and business governance agents to the AI-native ODA Canvas to validate actions, record decisions and support trusted autonomy at scale.

Building trust into every autonomous AI agent action
Communications service providers are moving beyond AI that assists people toward autonomous agents that reason, collaborate and take action across business and network systems. That shift creates a governance challenge: controls designed for people and predictable software do not provide a consistent way to determine whether an agent’s proposed action is secure, permitted and aligned with business intent before it happens, or to demonstrate afterwards why it was allowed.
Agents can communicate with one another and reach operational systems through tools, creating risks such as manipulated instructions, rogue behavior and data leakage. Governance may also remain manual and inconsistent across teams and vendors.
TM Forum Insights’ 2026 research, Trust and assurance: the keys to AI deployment at scale, conducted in partnership with IBM Institute for Business Value, found that 72% of CSPs believe their AI is trustworthy, but only 14% can provide evidence to prove it.
Without stronger assurance, autonomous AI can remain in pilots, while an incorrect decision could expose revenue, regulatory compliance and customer trust.
The Catalyst project, TrustFabric: Secure AI agents for telcos, proposes a standard way to evaluate every AI agent before it acts. Built into the TM Forum AI-native ODA Canvas, the approach is intended to help CSPs scale autonomous AI while retaining security controls, business accountability and a record of each decision.
TrustFabric adds two guardian agents to the AI-native ODA Canvas:
The two guardian agents are deliberately kept separate, a central feature of the Catalyst demonstration. One determines whether an action may happen from a security perspective, while the other decides whether it should happen from a business governance perspective. Each maintains its own record of its decisions.
The checks occur when an agent is created and whenever it proposes an action. Before acting, the agent explains its plan. The guardian agents can:
This keeps responsibility for revising the proposed action with the agent that acts, rather than with the guardian agents.
Every action is checked, and the operator chooses whether the check takes effect before the action runs or alongside it. This answers a common question about whether governance slows agents down.
The design also introduces a progressive trust model. Agents that behave consistently can earn greater freedom, while unusual behavior can be detected and stopped. Secure agent-to-agent collaboration and continuous governance are therefore treated as runtime capabilities rather than one-time approval exercises.
The Catalyst uses the following TM Forum assets and interfaces:
The Catalyst is developing a Security Agent and Business Governance Agent for the ODA Canvas. It also proposes contributions to GB1087 covering checks on agent plans before execution, trust scoring between agents and detecting rogue agents, alongside new guidance for the business governance of AI agents.
The project targets 100% AI action validation, continuous governance, and secure agent-to-agent collaboration. Intended business benefits include:
For the wider industry, the project proposes a ready-to-use trust model for agents from different vendors, built on TM Forum standards. For customers and society, the intended outcomes include fairer treatment, stronger protection of data, explainable decisions, and clear accountability to a responsible person.
“Trust is the foundation of every AI-native enterprise. TrustFabric shows how CSPs can securely scale autonomous AI through continuous governance, runtime validation, and trusted agent collaboration across the TM Forum ODA ecosystem,” said Asela Perera, Group Chief Information Officer at Dialog Axiata PLC.
TrustFabric addresses a barrier to autonomous operations that cannot be solved by technical performance. Agents must be able to act across systems, but operators also need a consistent mechanism for checking security, business purpose, compliance and accountability at the point of action.
By embedding guardian agents in the AI-native ODA Canvas and recording the reasoning behind each approval, escalation or block, the Catalyst offers a standards-based route from isolated pilots toward governed deployment. Its contribution is a reusable control pattern that can support autonomous AI without separating innovation from trust.
Meet the team at Innovate Asia in Bangkok, 17-19 November 2026.