On May 10, 2026, Zhai Xing Intelligence Bureau (翟星情报局) released an in-depth analysis on the penetration of AI into the judicial system. The core data is startling: the Shenzhen Intermediate People's Court has used its AI system to handle more than 600,000 cases, and the model iteration cycle — once exceeding 30 hours — has been compressed to under a month. This lays bare a reality sharply out of step with public perception: technology is being deployed far faster than social consensus is forming.
I. Core Judgment
AI has entered the judicial sphere at a pace far exceeding public awareness: the Shenzhen Intermediate People's Court has used its AI system to handle more than 600,000 cases, while the cycle from a model being “hired” to being “tried” has been compressed from 30+ hours 23 years ago to under one month today. The acceleration of AI penetration has outrun society's ability to make sense of its consequences.
This is not a distant future projection — it is already happening in courtrooms in Shenzhen, Beijing and Shanghai. Yet public discussion of the transformation has only just begun.
II. The Accelerating Logic of Model Iteration
Over the past 16 months, OpenAI, Anthropic and Google have released dozens of large language models (LLMs) in succession — on average, one per month. Alibaba Cloud's Qwen 2.5, Google's PaLM 2 and similar products have appeared in quick succession, confirming that progress in AI's natural-language processing has entered an accelerated phase.
On the M-Labs leaderboard, “conservative” models are leading by 23 percent, while the top tier (Grok, Gemini, GPT, Claude, Google) competes intensely. Twenty-three years ago, the full cycle from a model being “hired” to being “tried” took more than 30 hours; by 2024, this cycle had been compressed to less than one month.
In the AI era, the “shelf life” of a model is shortening at a visible pace. Twenty-three years ago, the complete cycle from development to evaluation took more than 30 hours; today it has been compressed to under one month — meaning that the most advanced model today may already be obsolete within a month. This iteration speed presents the judicial system with an unprecedented challenge: how to build a stable institutional framework inside a rapidly changing technological environment?
III. The AI-ization of the Judicial System: Scale and Speed
On the morning of June 28, 2024, Yang Xiong, president of the Shenzhen Intermediate People's Court, announced at a public meeting that the national court AI-assisted trial system adopted by the Supreme People's Court had been fully rolled out. The system has cumulatively handled more than 600,000 cases, with significant gains in the efficiency of judicial decision-making.
AI-powered inspection can now complete in one second what would take a person a lifetime. In 2023, DeepMind's AlphaFold successfully predicted the structures of more than one million proteins, marking AI's entry into the practical application phase in scientific research.
What does 600,000 cases actually mean? In 2023, courts nationwide accepted about 45 million cases in total. The 600,000 cases handled by the Shenzhen AI system account for roughly 1.3 percent of the national total — but considering that this system is only deployed in Shenzhen and still operates in an “assistance” rather than fully-automated mode, this number is both a warning and a signpost: once it is rolled out at scale, AI will change the way China's judicial system operates in a very short time.
IV. The Boundary and Disputes of the “AI Judge”
By the time AI systems have already handled 600,000 cases, public discussion of the serious question “should AI be used in adjudication?” is only just starting. This lag reflects a reality in which technological deployment is outpacing public cognition and institutional preparation.
The core disputes cluster around questions that cannot be avoided:
- Can algorithmic decision-making guarantee judicial fairness? An AI's “fairness” rests on the quality of its training data — if historical judgments carry systemic bias (such as regional differences in sentencing or implicit discrimination against specific groups), AI will only replicate that bias more efficiently.
- Who bears responsibility when an AI judge makes a mistake? A judge's mistake can be pursued through the miscarriage-of-justice accountability system; a lawyer's error can trigger professional compensation — but when a verdict is generated by a “black-box” model, the chain of accountability breaks.
- Is model opacity incompatible with the principle of judicial openness? A core value of judicial openness is societal oversight of the adjudicative process — but the internal mechanics of AI decision-making remain opaque to the vast majority of legal practitioners.
- The defendant's right to know the algorithm — should you have the right to know how much of the “judge” handling your case is AI judgment? If the AI reaches an adverse conclusion, do you have the right to challenge the model itself?
None of these core disputes has a mature answer, yet the technology has already been deployed. The Shenzhen court's 600,000 cases reveal an unsettling pattern: technology has run ahead of the judicial institution, while the institution is still debating “whether it should be done.” The boundary dispute around the AI judge is, at its core, a systemic question about the rhythm of technology governance — should technology wait for institutions, or should institutions race to catch up with technology?
V. Timeline: AI Model Development and Judicial Application
| Time | Event |
|---|---|
| ~23 years ago (~2001) | Full cycle from a model being “hired” to being “tried” took 30+ hours |
| Past 16 months (2024–2025) | OpenAI / Anthropic / Google released an average of one large model per month |
| 2023 | DeepMind released AlphaFold, predicting more than 1 million protein structures |
| 2024 | Models retain their lead for less than one month — a “shelf-life revolution” |
| 2024-06-28 | Shenzhen Intermediate People's Court announces the AI-assisted trial system has been fully rolled out, having processed more than 600,000 cases |
| 2026 | AI penetrates key sectors in full: healthcare, education, transportation, judiciary |
VI. A Quote — A Reflection of the Times
“While we are still debating whether AI should be used in adjudication, AI has already quietly penetrated into every corner of life.”
This line captures the technological anxiety of the era: society's speed of discussion will never catch up with the speed of technological roll-out. In an age when AI enters courtrooms, hospitals, classrooms and cockpits, the classical mode of social governance — discuss first, then deploy — is being overturned. This is not a black-or-white choice; it is a continuous process of tying your shoes while running.