Law report No. GLW-3473 · filed November 17, 2026
Legal PracticeReported case
Crosby Secures $5m Lloyd's-Backed Insurance Cover for AI Paralegal Agents
New-model firm Crosby has a $5m Lloyd's-backed policy from Testudo covering its AI paralegal agents — believed to be the first such cover for law firm AI outputs.
By Amara Osei4 min read855 words
Holding
- Crosby secured a $5m insurance policy, effective from the start of November, covering AI paralegal agent outputs — believed to be a first for a law firm
- The policy is provided by Testudo and underwritten by Lloyd's of London
- Agents reached 94% accuracy on escalation before initial launch
- Human review time fell to zero for 20% of pre-existing legal tasks, saving 36 of roughly 180 daily review hours

A $5 million insurance policy took effect at the start of November, covering the work of AI paralegal agents at new-model law firm Crosby — a policy the firm believes is the first liability cover any law firm has obtained for the outputs of its AI agents. The cover comes from Testudo and is underwritten by Lloyd's of London.
The arrangement marks a structural shift in how legal work is delegated. Crosby, which describes itself as an AI-native law firm, operates paralegal agents as what it calls an "agentic front door" for client requests before matters reach human lawyers. Every new client message passes through an agent.
How do the agents work?
Each agent can complete end-to-end work within a defined capability set, drawing on up-to-date context that Crosby stores on each client. For tasks outside that set, the agent passes documents and context to human lawyers and communicates updates back to clients in real time.
The firm measures every new capability on dedicated datasets. Before its initial launch, Crosby reached 94% accuracy on escalation — measured by whether the agent correctly completed the task itself or correctly escalated it to a lawyer. For each new feature, Crosby builds datasets and judges over agent trajectories to measure individual task performance.
Security controls sit alongside the accuracy metrics. The agents work in a sandboxed environment where they can read and write documents, with access limited to the relevant client's business data. The firm reports total data isolation between clients. Trust, in this model, is engineered rather than assumed.
What does the insurance actually cover?
The $5m policy, effective from the start of November, covers two categories of agent activity:
- Agents handling legal tasks under the supervision of an attorney
- Agents performing paralegal work on their own
Securing the cover was not a formality. Testudo conducted a technical analysis of Crosby's systems before underwriting, examining:
- The technology and models underlying the platform
- Governance controls and logs of human review steps
- How the agents work, what data the systems access, and how the models are trained
- Internal company contracts and policies
Testudo collected further data to determine the firm's AI liability exposure. The underwriting process, in effect, treated an algorithmic workforce the way an insurer would treat any other source of professional risk.
CEO Ryan Daniels and the Crosby team supplied the detail in response to questions. In plain terms: the firm has bought peace of mind for its machines.
How much lawyer time do the agents save?
The numbers are concrete. After the client agents launched, human review time fell to zero for 20% of Crosby's pre-existing legal tasks — work such as summaries of negotiations, pulling prior business context, and routing matters to relevant parties for approval.
Typical human review time across the firm runs at roughly 180 hours per day. A 20% reduction therefore saves 36 hours of lawyer time every day — the equivalent of adding several full-time lawyers to the roster without hiring anyone, or reclaiming more than 900 billable-adjacent hours a month.
The agents also do work lawyers never did. Crosby notes that clients will soon be able to ask agents questions they would never put to a lawyer — for example, the last time the client agreed to a 3x liability cap. The agents, the firm says, have far better memory and respond in under five seconds.
Is delegation of legal responsibility to AI now insurable?
The industry has debated delegating responsibility to AI for some time. Crosby's arrangement does exactly that, albeit with lawyer oversight of final outputs. Much of the work, however, now sits with the agents. That, in short, is why the insurance exists.
The significance extends beyond one firm. If the new-model approach to legal work is to scale, agent delegation becomes increasingly important — and firms need to know their agents will perform correctly, just as a partner must trust an associate. Crosby's answer is to treat the agent literally like a paralegal, complete with professional indemnity.
Improved technology and better "harnesses" — the scaffolding firms build around generative AI — will help. But the insurance component raises the question to a different level: it converts a management bet into a transferable, priced risk.
The timing invites perspective. It is now four years since ChatGPT arrived, and a firm has already underwritten its AI workforce at Lloyd's. Anyone who doubts that this approach will be mainstream across law firms, new-model outfits and in-house teams by 2030 is, on the market's current direction of travel, betting against the flow.
What should practitioners take from this?
For law firm risk and innovation leads, the Crosby policy offers a template. Insurers will now underwrite AI agent output — provided the firm can demonstrate governance controls, human review logs, data isolation and a documented technical basis for its models. Firms deploying agents without such evidence may find both insurers and clients asking harder questions.
The practical consequence: AI delegation in legal services now carries a price, a policy number and an underwriter. That is what legitimising looks like.
via legalinnovatorsnewyork.com (Original)
More from Amara Osei
Show full bio
Senior reporter covering people, launches and day-to-day practice in International law and legal practice at Global Law Wire.
327 articles
Also before the court
- A&O Shearman Commits $1.1 Billion and Billable Hours to AI Build-Out
- Ex-Legora Founders Raise $2.4m for AI-Native Real Estate Law Firm Athena
- China Issues First Dedicated Civil-Liability Rules for AI Disputes
- Robot Arbitrators Spark Conflicts Over AI in Dispute Resolution
- Addleshaw Goddard and Fieldfisher Split Over Rival Legal AI Giants Harvey and Legora