Yixin (02858) announces open-sourcing of the Harness governance system to solidify the compliance foundation for long-horizon AI agents in auto finance

On November 5, at the 2026 World Internet Conference Wuzhen Summit, AI-driven fintech platform Yixin (02858) officially announced the open-sourcing of its self-developed Harness system. This marks the second time Yixin has opened core AI capabilities to the industry, following its 2025 open-sourcing of the industry's first Agentic large model, signaling that its AI open-source strategy has advanced from "opening up model capabilities" to "sharing the governance foundation" providing a reusable steering framework for the scaled application of agents in highly regulated industries such as auto finance. Jia Zhifeng, Chief Technology Officer of Yixin, speaking at the Wuzhen Summit The World Internet Conference Wuzhen Summit is the largest and highest-level internet conference held in China. With the theme "Openness and Cooperation for a Shared Future Jointly Building a Community with a Shared Future in Cyberspace," this year's conference is an important window for observing frontier trends in the global internet and artificial intelligence. Yixin's choice to announce the open-sourcing of the Harness system here is both an active response to the industry's open ecosystem and a demonstration to the world of the technological strength of China's auto finance AI. In a forum speech at the Wuzhen Summit, Jia Zhifeng, Chief Technology Officer of Yixin, said that auto finance is a typical long-chain, complex decision-making scenario. From customer onboarding, document submission, approval, plan matching, delivery and signing, to asset management, it is characterized by long task spans, numerous process nodes, difficult dynamic decision-making, and strict compliance controls. To truly adapt to industry needs, Yixin has built a new AI paradigm the "Human-Centric Long Horizon Agent." The core logic of this paradigm is to operate around the real needs of "people": a human understanding model continuously captures the customer's car-purchase intent, budget level, and decision stage; vertical domain models match dynamic service strategies; and the Harness system coordinates state management and compliance control across the entire process. Working together, the agent is no longer a tool that mechanically executes instructions, but a long-term service partner capable of adjusting the pace of service in real time as the customer's state changes maximizing business conversion efficiency and user service experience while strictly adhering to compliance bottom lines. The establishment of this paradigm requires not only the quality and efficiency improvements of the Agentic large model, but also the support of the Harness governance system. The Harness steering system built by Yixin uses five types of management mechanisms memory, state, belief, goals, and strategy to achieve continuous tracking and dynamic decision-making across the customer's full lifecycle. It also establishes a human-machine collaboration mechanism that, in the event of compliance risks, major changes in rights and interests, or special exception scenarios, can trigger circuit-breaking and governance strategies at millisecond-level and switch to a manual workflow, ensuring that every transaction runs smoothly within a safe and controllable scope. At present, the "Human-Centric Long Horizon Agent" AI system has been deployed at scale across Yixin's entire business chain. As of the end of June 2026, Yixin had deployed 45 agents in real business scenarios, covering core areas such as marketing, risk control, approval, and customer service, processing an average of 230,000 tasks per day, with agents independently delivering 90% of business, driving a 45.4% increase in overall operational efficiency. Leveraging its "long-horizon agent" capabilities, Yixin's AI can independently follow up with the same customer for more than 20 days, continuously execute a single task for more than 16 hours, conduct up to 81 rounds of dialogue in a single session with more than 400 message interactions throughout, and has shortened the average time from lead contact to transaction to 3 days. At the Wuzhen Summit, visitors experience Yixin's AI Agent Jia Zhifeng said that the open-sourcing of the Harness system aims to lower the barrier to deploying agents in auto finance and similar highly regulated industries, enabling more institutions to quickly build safe, compliant, and efficient AI application systems without having to build governance frameworks from scratch. Going forward, Yixin will continue to deepen the full-scenario deployment of the "Human-Centric Long Horizon Agent," empower upstream and downstream industry partners with an open and open-source posture, and jointly drive the high-quality development of the auto finance industry toward greater intelligence, standardization, and inclusiveness.