What is Agent Trust Score?

The Problem: Autonomous AI agents are becoming economic actors. They trade, they negotiate, they execute. But who rates their trustworthiness? Who tells you if an agent will honor its commitments, stay within regulatory boundaries, and behave predictably?

The Answer: Agent Trust Score — a scene-specific rating system that provides regulatory context for autonomous AI agents operating in APAC financial markets.

Why Agents Need Ratings

As AI agents become more autonomous, three trust problems emerge:

1. Counter-party Risk: If I'm negotiating with an AI agent, how do I know it will honor the deal? Does it have a track record?
2. Regulatory Compliance: Will this agent's actions trigger regulatory issues? Is it operating within allowed boundaries?
3. Predictability: Can I predict how this agent will behave in edge cases? Has it been tested in similar scenarios?

Today, there's no standard way to answer these questions. Every AI product company reinvents compliance checks. Every agent is a black box to its counter-parties.

The "Agent Moody's" Vision

Credit rating agencies (Moody's, S&P, Fitch) solved a similar problem for financial entities. Before ratings, lenders had to individually assess every borrower. Ratings created a standardized trust layer.

Agent Trust Score brings this infrastructure to autonomous AI:

Scene-Specific Ratings

Regulatory environments differ dramatically across APAC jurisdictions. An agent operating in Hong Kong faces different rules than one in Singapore or Japan. Agent Trust Score provides context-aware ratings:

How It Works

Agent Trust Score operates on a two-phase mechanism:

Phase 1: Context Check (Before Action)

Before an agent executes an action, it (or its operator) queries the Trust Score API to understand the regulatory context. The API returns:

• Trust Score — A letter grade (AAA to D) indicating overall trustworthiness for this scene
• Confidence Level — How certain we are about the rating (based on rule clarity and track record)
• Flags — Specific regulatory considerations (Socratic questions, not answers)
• Track Record — Historical performance data for this agent in similar scenes

Phase 2: Outcome Reporting (After Action)

After the action completes, operators report the outcome. This builds the agent's track record:

• Transaction completed successfully?
• Any disputes or incidents?
• Regulatory feedback received?

This creates a feedback loop that improves rating accuracy over time.

No Conflicts of Interest

Traditional credit ratings have a fundamental conflict: the rated entity pays for its rating. This creates incentive misalignment that contributed to the 2008 financial crisis.

Agent Trust Score takes a different approach:

Aspect	Traditional Ratings	Agent Trust Score
Who pays?	Rated entity (issuer-pays model)	Querying party (subscriber-pays model)
Conflict?	Yes — incentive to inflate ratings	No — rated agents don't pay us
Transparency	Methodology often opaque	Public track record, open post-mortems
Accountability	Limited liability for wrong ratings	Failed predictions publicly disclosed

We charge the parties querying trust scores, not the agents being rated. This ensures our incentive is accurate ratings, not happy customers who want inflated scores.

The Socratic Approach

Agent Trust Score doesn't tell you "yes" or "no". It asks the right questions.

This is intentional. Regulatory compliance ultimately requires human judgment. Our role is to surface the relevant considerations — not to replace compliance teams.

Example flags:

• "SFC may view this as a Type 1 regulated activity — have you confirmed your license coverage?"
• "Similar transactions in Singapore required prior MAS notification — does HK have equivalent requirements?"
• "This agent's dispute rate (2%) is higher than market average (0.5%) — acceptable for your risk profile?"

We flag risks. You make decisions. This keeps the accountability where it belongs.

Who Needs This?

AI Product Companies

Building agents that operate in financial contexts? You need compliance guardrails that work at agent speed. Agent Trust Score provides pre-built regulatory context so you don't reinvent compliance from scratch.

Financial Institutions

Partnering with or deploying AI agents? You need to assess agent risk before integration. Agent Trust Score gives you standardized ratings to compare agents and monitor ongoing risk.

Agent Developers

Building an agent that needs to be trusted? Agent Trust Score lets you build a verifiable track record. Good behavior gets recognized; bad actors get flagged.

Current Scope: APAC

Agent Trust Score currently covers regulatory contexts across Asia-Pacific:

• 🇭🇰 Hong Kong — SFC, HKMA, VASP regime
• 🇸🇬 Singapore — MAS, PSA framework
• 🇯🇵 Japan — FSA, JFSA regulations
• 🇰🇷 South Korea — FSC, DABA framework
• 🇦🇺 Australia — ASIC, AUSTRAC
• 🇹🇭 Thailand — SEC Thailand
• 🇮🇩 Indonesia — OJK, Bappebti
• 🇵🇭 Philippines — BSP, SEC Philippines

More jurisdictions coming. Global expansion planned for 2027.

What's Next

We're building Agent Trust Score with these principles:

1. Start with context, not judgment — Regulatory scene awareness first, ratings build from there
2. Track record over promises — Actual outcomes matter more than marketing claims
3. Open post-mortems — When we're wrong, we explain why publicly
4. No conflicts — Queriers pay, rated agents don't

The agent economy is coming. Trust infrastructure needs to be ready.