Legal AI liability insurance pays a maximum of 50,000 yuan per incident; lawyers explain who is liable when AI makes a mistake.
In August 2026, PowerLaw AI and PICC P&C Beijing Branch launched artificial intelligence service liability insurance for the Wulv AI Lawyer Agent, covering specific losses caused by risks such as large-model hallucinations, false information, and misleading answers, with a maximum payout of RMB 50,000 per AI liability occurrence. Almost at the same time, U.S. legal practice saw a series of incidents in which courts sanctioned AI-fabricated case citations. On one side, the risks are beginning to materialize; on the other, they are beginning to be underwritten—the industry agenda for legal AI is moving from capability to liability.
As a legal team that has long served corporate compliance and technology law matters, we are more concerned with this: when AI gives incorrect legal answers and causes losses, how is liability allocated among foundation model providers, legal AI platforms, lawyers, and clients? Under the current legal framework, how should each party plan ahead for compliance obligations?
If a legal AI makes a mistake, what legal liabilities might be triggered?
Legal AI is fundamentally different from ordinary tool software. What it outputs is not formatting or data reports, but legal opinions that carry the nature of professional judgment. Once an error occurs, it may simultaneously trigger multiple legal liabilities.
From the perspective of contractual liability, when a user purchases legal AI services, a service contract relationship is established between the parties. If the platform advertises that it can "review contracts like a lawyer," but due to model hallucination it omits key clauses and causes damage to the client's transaction, the user may assert liability for breach of contract under Article 577 of the Civil Code. In practice, the focus of dispute lies in whether the platform has adequately indicated in the user agreement disclaimer clauses such as "for reference only and does not constitute legal advice," and whether such clauses have fulfilled the conspicuous notice obligation under Article 496 of the Civil Code.
From the perspective of tort liability, if erroneous information output by AI causes a user to suffer pure economic loss, the user may attempt to assert fault-based tort liability under Article 1165 of the Civil Code. However, tort remedies for pure economic loss have always been approached cautiously in China's judicial practice, and the plaintiff must prove that the platform was at fault and that a causal relationship exists between the loss and the AI output, making the burden of proof considerable.
From the perspective of professional responsibility, if a lawyer uses AI tools and submits the output to a court or client without verification, the lawyer must still bear the corresponding professional liability under Article 49 of the Lawyers Law. The Shanghai Bar Association's 2026 "Operational Guidelines for AI Application in Lawyers' Corporate Legal Advisory Services (Trial)" explicitly requires lawyers to conduct manual review of AI output, including factual accuracy, the authenticity and validity of legal bases, legal logic, risk identification, and the feasibility of recommendations. The American Bar Association's Formal Opinion 512, issued in 2024, likewise emphasizes that lawyers using generative AI must still fulfill existing professional obligations such as competence, confidentiality, communication, supervision, and candor to the tribunal.
This means that AI can assist professional work, but it cannot automatically replace lawyers in assuming final professional judgment.
What can liability insurance solve, and what can it not solve?
The emergence of legal AI liability insurance essentially turns "AI risk" from an abstract concept into a calculable unit of risk. The fact that insurance companies are willing to underwrite it shows that the market has begun trying to price AI errors. But a limit of 50,000 yuan carries vastly different implications across different legal scenarios.
For low-risk scenarios such as contract information organization, basic legal consultation, and review of standard-form clauses, 50,000 yuan may offer some meaningful compensation. However, in high-risk matters such as determining the statute of limitations, major mergers and acquisitions, and compliance with mandatory regulatory obligations, a single error could lead to losses of tens of millions of yuan or even more. The currently available public information does not yet indicate whether tiered underwriting is applied to different legal tasks, and the actuarial basis behind the 50,000 yuan cap also remains to be seen.
More critically, insurance addresses the sharing of economic risk and cannot replace the determination of underlying legal liability. A legal AI service may involve foundation model providers, legal AI application platforms, data and knowledge base suppliers, lawyers or law firms, corporate clients, and insurance institutions. A single error may simultaneously trigger contractual liability, tort liability, lawyers' professional liability, and platform governance obligations. After an insurance claim is paid, the insurer may also exercise subrogation rights against a party at fault.
Therefore, when configuring risk protection, a more realistic approach for enterprises or law firms is a combined solution: law firm professional liability insurance plus technology and cyber risk coverage, further supplemented by specific AI-related coverage, with layered configuration based on different causes such as lawyers failing to verify AI output, platform service defects, data breaches, and intellectual property disputes.
The compliance obligations of generative AI cannot be avoided by either platforms or law firms.
China's Interim Measures for the Administration of Generative AI Services establish obligations based on a basic structure of "service providers—service users," requiring providers and users to sign service agreements, clarify the rights and obligations of both parties, and specify and disclose the applicable user groups, occasions, and purposes of the service. Where laws and administrative regulations provide that providing relevant generative AI services requires administrative licensing, providers shall obtain a license in accordance with the law.
The Measures for the Labeling of AI-Generated Synthetic Content further set out requirements for explicit labels, implicit labels, dissemination prompts, user agreements, and log retention under specific circumstances. Although these rules do not directly answer how losses caused by erroneous legal advice should be allocated, they at least show that generative AI is not a technological product existing in a responsibility vacuum.
For law firms, AI applications running ahead of governance systems is a real risk. Thomson Reuters Institute's 2026 legal industry research shows that 34% of professionals report using AI tools not authorized by their organizations. When lawyers privately use AI tools not approved by their firms to handle client materials, they may cross multiple compliance red lines, including confidentiality obligations, cross-border data transfer, and conflict-of-interest screening. Law firms should establish an AI tool access list, usage approval processes, output verification mechanisms, and an incident documentation system as soon as possible, integrating AI applications into their existing legal practice risk management systems.
For legal tech companies, the more a product emphasizes that it "works like a lawyer," the more it needs to explain the relationship between disclaimers such as "for reference only" and "does not constitute legal advice" and its service promises. If there is a gap between promotional language and actual service capabilities, it may constitute false or misleading commercial advertising prohibited by Article 8 of the Anti-Unfair Competition Law.
From a lawyer's perspective, the real value of AI liability insurance lies in driving governance standards.
Once insurance institutions bear real claims risk, they have an incentive to require policyholders to provide more structured risk evidence: whether legal data sources are stable, whether statutes and cases are updated promptly, whether model outputs can be traced, whether human gates are set for high-risk tasks, whether models are retested after upgrades, and whether anomalous events are fully documented. Once underwriting and renewal rules stabilize, they may in turn push legal AI toward more unified testing, auditing, and governance standards.
From this perspective, insurance companies may become a new type of "external quality inspector" in the legal AI ecosystem. For legal tech companies, establishing verifiable data governance, model monitoring, and human review processes as early as possible is not only a compliance requirement but also a prerequisite for obtaining insurance coverage and customer trust in the future. For law firms and corporate legal departments, when procuring legal AI services, they should make liability clauses, data security, output verification, incident response, and insurance coverage the core elements of contract negotiations, rather than focusing solely on feature demonstrations and pricing.
Guangdong Zhiming Law Firm has long focused on the intersection of artificial intelligence and legal services. We provide legal tech companies with product compliance reviews, user agreement design, and layered liability risk solutions, and we also help law firms and corporate legal departments build AI tool admission systems, data compliance frameworks, and professional risk prevention services. When AI begins to take on professional judgment functions, the value of legal professionals lies not in rejecting technology, but in defining the boundaries of liability for it.
