Bloomberg Reports AI Is Forcing Big Law to Rethink Business Model as Kirkland Hints at Fine-Tuning Open-Source LLMs for Proprietary Legal AI
A Bloomberg feature published today examines how artificial intelligence (AI) is forcing major law firms to fundamentally rethink their business models, from billing structures to staffing ratios and technology investment. Separately, Kirkland & Ellis — which committed $500 million to build a proprietary AI platform — has signalled through its hiring activity that its project may involve fine-tuning open-source LLMs (large language models, the technology underlying tools like ChatGPT) on its own legal data, rather than simply licensing existing commercial platforms. Two newly posted AI Infrastructure Director roles at Kirkland require experience with 'on-premise GPU environments' — specialist computing hardware used to train and adapt AI models locally rather than via the cloud — indicating the firm may be building a model tailored to its specific practice needs. Kirkland's job adverts also reference familiarity with existing legal AI tools including Harvey, Legora, CoCounsel, and Lexis+ AI, suggesting the firm is using commercial platforms as a benchmark while building a proprietary alternative. The Bloomberg piece frames this as part of a broader industry reckoning: as AI absorbs more routine legal work, firms are being forced to decide whether to buy, build, or partner on AI infrastructure — a choice with significant implications for firm economics and junior lawyer pipelines.
Why this matters
The decision by Kirkland & Ellis to potentially fine-tune proprietary open-source LLMs rather than rely on commercial legal AI platforms represents a significant escalation in the 'build vs buy' debate in Big Law. Fine-tuning a model on firm-specific data creates competitive differentiation but also raises acute questions around data governance, client confidentiality, and technology licensing — all areas requiring specialist legal input. The Bloomberg framing — that AI is forcing firms to rethink 'business as usual' — captures a structural shift in how legal services are produced and priced that will affect trainee hiring, billing models, and client expectations simultaneously. For City firms, the Kirkland model sets a benchmark: either invest at comparable scale or risk being outpaced on efficiency and margin.
On the Ground
A trainee advising a law firm or legal tech vendor on AI infrastructure would review technology licence agreements for open-source LLM terms, mark up data processing agreements to address confidential client data use in model training, and draft AI governance policy documentation for internal use. They might also assist with vendor due diligence questionnaires for AI tool suppliers.
Interview prep
Soundbite
Firms that build proprietary LLMs on their own data create a moat — but client confidentiality in training data is the legal tripwire.
Question you might get
“What legal and ethical issues arise when a law firm uses its archive of client documents to fine-tune a proprietary AI model, and how should those risks be managed?”
Full answer
Kirkland & Ellis is hiring AI Infrastructure Directors with on-premise GPU experience, strongly suggesting its $500 million AI project involves fine-tuning open-source large language models on proprietary legal data rather than licensing off-the-shelf tools. This matters because a firm-specific AI model trained on years of deal documentation could dramatically compress the time for drafting, due diligence, and research — shifting the economic case for large associate pools. The wider trend, captured in Bloomberg's feature today, is that AI is forcing every major firm to make irreversible capital allocation decisions about technology strategy. I think the firms that build proprietary models will hold a durable advantage on high-volume, data-intensive work — but the client confidentiality and data governance questions in the training process are genuinely unresolved and will require careful legal structuring.
Sources
My notes
saved