Courts Draw a Line Between Trade Secrets and Corporate Accountability
China’s Supreme People’s Court has issued new guidance on handling disputes involving artificial intelligence, and the core message is blunt: a company’s algorithm may remain a black box, but the company itself cannot. The opinion, summarized in a widely circulated legal commentary, lays out how courts should approach AI-related litigation — from liability for model outputs to the burden of proof when a plaintiff cannot inspect the underlying system.
What the Guidance Actually Changes
The practical effect is a shift in evidentiary expectations. Under the emerging framework, AI developers cannot simply invoke “model opacity” or trade-secret protection to shut down discovery. Courts are being directed to examine:
- Training data provenance — whether datasets were lawfully obtained and whether copyright or personal-information violations occurred during collection.
- Deployment and supervision — whether the operator had reasonable guardrails, logging, and human review in place.
- Harm causation — how liability is apportioned among model developers, platform operators, and end users when an AI system causes damage.
- Disclosure obligations — what a defendant must reveal about model architecture, even if the full weights remain confidential.
The guidance does not force companies to open-source their models. It does, however, erode the idea that “the algorithm did it” is a complete defense. If a firm deploys an AI system commercially, it is expected to be able to explain how that system was built, tested, and monitored.
Why This Matters Beyond China
For AI startups, the implications are operational, not just legal. Compliance teams will need audit trails, model cards, and documented data lineage long before a lawsuit arrives. Investors conducting due diligence will start asking for the same. The cost of building an AI company is rising — not because compute is expensive, but because accountability is now a line item.
The crypto and Web3 sector should pay particular attention. Decentralized AI projects that tout “trustless” or “permissionless” model deployment face a harder question: if the model is a black box and the operating entity is anonymous or offshore, who is liable when it causes harm? The Supreme Court’s logic suggests that courts will look for a responsible legal person behind the code. Projects structured to avoid that accountability may find themselves at a disadvantage in any jurisdiction that adopts similar reasoning.
The Road Ahead
Expect this guidance to be cited in commercial disputes, IP cases, and personal-injury claims involving AI outputs. It also sets a template that other regulators may borrow: you can keep your model proprietary, but you must be able to demonstrate that you understood and controlled its risks. For AI founders, the era of plausible deniability is closing. For lawyers and compliance officers, the work is just beginning.
