Overview
AI agents are moving from passive response to active execution, but rising autonomy is pushing identity, authorization and accountability to the center of regulation and corporate agendas. On November 2, the Hong Kong Monetary Authority launched the SAGE project at Fintech Week, targeting agentic shopping payments, corporate treasury and machine-to-machine trading, while studying a trusted-agent registry and Know Your Agent (KYA) standards with a cross-sector expert panel.
The same day, Goldman Sachs—after weighing productivity and risk—barred most staff from advanced coding AI tools (Claude Code, GitHub Copilot), limiting them to engineers. Tencent’s Hunyuan released Hyra 2.0, a research-agent framework letting general agents build reusable scientific models and tools, even producing mathematician-verified conjecture proofs.
Background & Interpretation
1. Technical Background and Evolution
The agent narrative is entering a trust phase. As agents autonomously call tools, run multi-step tasks and trade on behalf of users or machines, trusted identity, bounded authorization and traceability become prerequisites. Wuzhen’s agent-interoperability track echoes this turn.
2. Core Drivers and Underlying Mechanism
KYA mirrors KYC—registering agent identity, defining permissions and auditing behavior to answer who acts on my behalf. Enterprises tightening tool access hedge data-leak and code-quality risks. The two ends converge to pave the way for agent scale-up.
3. Market Structure and Value-Chain Impact
Research agents first reach high-value scenes: Hyra 2.0 lifted the Xiangshan RISC-V processor’s per-GHz performance by 5.79% and produced reusable tools, showing agents moving from solving problems to building tools. Finance, law and R&D will be the first battlegrounds for trusted agents.
Implications & Outlook
Near-term, regulatory frameworks (KYA, registries) and enterprise permission governance advance in parallel; medium-term, trusted agents may unlock productivity in controlled settings. Watch SAGE pilot progress, major banks’ AI governance stances, and standardization of agent auditing.
Takeaways for Industry Participants
Financial institutions should build agent identity/permission frameworks early; software firms can pilot coding agents in controlled environments with quality gates; regulators should accelerate interoperable, traceable standards to cut cross-domain friction.
This column compiles industry information and shares technical perspectives; it does not constitute investment advice.