New Milestone in Legal AI: Liability Insurance Now Available for AI Agents' Work
A U.S. law firm has secured a $5 million liability insurance policy for work performed by its AI agents. Before signing the contract, the insurer not only reviewed the technology but also audited the firm's control processes and documentation of human reviews. This case could represent a significant advancement in the application of artificial intelligence in legal practice.
Artificial intelligence is increasingly capable of taking over tasks in the daily operations of law firms. It analyzes documents, reviews contracts, retrieves information, and, with proper integration, can even independently execute various administrative operations.
However, as AI assistants evolve into more autonomous AI agents, a fundamental question arises: who bears the responsibility if the AI makes a mistake?
A poorly prepared document, a missed deadline, or a misinterpreted client requirement can have legal and compensatory consequences, regardless of whether the task was performed by a human or artificial intelligence.
The Crosby, a U.S. law firm employing an AI-based operational model, has now provided a noteworthy answer to this question.
$5 Million Insurance for AI Agents' Work
In November 2026, Crosby announced that it had secured a $5 million liability insurance policy specifically for work performed by its AI agents.
The insurance is provided by Testudo, with the risk underwritten by Lloyd’s of London.
The coverage extends to two types of activities:
- Legal tasks performed by AI agents under attorney supervision.
- Legal assistant tasks performed independently by agents within their defined scope.
Crosby claims this is the first instance where a law firm has secured liability insurance specifically for the outcomes of its AI agents' work.
This development is particularly significant because the insurance is not merely an extension of traditional professional liability insurance. Risks arising from the autonomous activities of AI agents have also been included in the range of risks assessed and undertaken by the insurer.
What Did the Insurer Examine Before Signing the Contract?
The insurance agreement was preceded by a detailed technological and operational audit.
Testudo examined the law firm's AI systems, the models used, data access rules, internal control mechanisms, and workflow documentation.
The audit extended to logs of human reviews, internal policies, and contracts.
Thus, the insurer did not simply evaluate what tasks the applied artificial intelligence could perform but also how the firm monitors, restricts, and documents the activities of AI agents.
This is an important distinction.
The technological capabilities of an AI system alone do not guarantee reliable operation. Equally important is what data it can access, what operations it can perform, when it must seek human approval, and how its activities can be reconstructed afterward.
How Do Crosby's AI Agents Operate?
In Crosby's operational model, every incoming client message is initially processed by an AI agent.
The agent determines, based on available client data and predefined permissions, whether it can independently perform the requested task.
If so, it executes the task. If the request falls outside its scope, it forwards it to an attorney with the necessary documents and the full context of the case.
Clients are clearly informed when they are communicating with an AI agent.
For data security, agents operate in a segregated, restricted environment. They can only access business data related to the specific client, and each client's data is handled separately.
Thus, the system not only answers questions but also performs actual tasks within defined limits.
What Results Have Been Achieved with AI Agents?
According to data provided by Crosby, the use of AI agents has already resulted in significant changes in the firm's work organization.
94% Escalation Accuracy
In pre-implementation tests, AI agents accurately determined 94% of the time whether a task could be performed independently or needed to be forwarded to an attorney.
This is not a general accuracy indicator for all legal tasks but specifically measures the effectiveness of decision-making between task execution and proper forwarding.
20% Reduction in Tasks Requiring Human Review
For approximately 20% of tasks previously performed by attorneys, the time spent on human review has been reduced to zero.
These typically include activities such as summarizing negotiations, retrieving previous business information, or forwarding documents to the appropriate approvers.
According to the firm's calculations, a total of 36 attorney work hours are saved daily.
Response Time Within 5 Seconds
The system can respond within five seconds for certain information retrieval tasks.
For example, determining what liability limitation terms were applied in previous contracts with a given client.
These tasks previously required longer document searches and manual checks.
However, it is important to note that the results are based on the firm's own disclosures and cannot be considered general performance indicators applicable to all law firms using AI agents.
Insurance Does Not Replace Attorney Responsibility
Crosby's case does not mean that AI agents can take over legal tasks without restrictions or that liability insurance alone exempts from attorney oversight obligations.
Insurance coverage, technological autonomy, and professional responsibility are three separate issues.
The range of activities that AI agents can perform independently is still determined by applicable laws and professional regulations. Liability insurance can provide coverage for damages incurred under the terms of the contract.
Insurability is not the same as AI performing legal activities without restrictions and attorney control.
The real significance lies in the fact that risks arising from the autonomous operation of artificial intelligence can now be assessed in a form that an insurer is willing to financially cover.
Why Is a Proper Administrative Background Becoming More Important?
The introduction of AI agents can elevate the operation of law firms to a new level.
As long as artificial intelligence only drafts texts or summarizes documents, the attorney typically directly reviews and uses the result.
The situation changes when AI can independently access cases, manage documents, create tasks, modify data, or communicate with clients.
In such cases, the verifiability of AI activities becomes not only a technological but also an administrative issue.
In a well-designed operational environment, it is precisely defined what an AI agent:
- Can access in terms of cases, documents, and client data.
- Can perform independently.
- Requires attorney approval for.
- How completed operations are recorded and logged.
- How a particular activity or decision can be retrospectively verified.
These principles have always played a crucial role in the operation of a well-organized law firm. However, with the advent of artificial intelligence, they gain even greater importance.
Orderly case and document management, a transparent authorization system, and traceable workflows not only support traditional legal work but also provide a foundation for the safe application of AI integrations.
Could a New Era Begin in Legal Work?
Crosby's example clearly shows that the application of artificial intelligence in legal practice can gradually go beyond document generation and simple information retrieval.
AI agents are no longer just tools supporting the work of attorneys but also appear as partially autonomous entities with defined permissions and roles.
Consequently, the organization of legal work must also adapt to new possibilities.
The question is increasingly less about whether artificial intelligence can perform certain tasks. It is more about under what conditions, with what control mechanisms, and within what liability frameworks these tasks can be delegated.
The insurance secured by Crosby does not answer all emerging legal and professional questions, but it indicates that risk management related to AI agents' activities has already appeared in the insurance market.
The foundation for responsibly applicable AI is not only the appropriate technology but also the well-regulated and verifiable operational environment in which the technology operates.
For law firms, this could be one of the most important considerations in planning AI developments in the coming years.
Sources:
- Artificial Lawyer – Crosby Insures Paralegal Agents – Interview (November 10, 2026)
- Crosby – AI Agent Insurance: Why Crosby Insured Its Legal Agents (November 9, 2026)

Zoltán Kéri