Q&A with David Wakeling of Allen & Overy, one of world's largest law firms, on using OpenAI-developed generative AI tool Harvey to write legal documents
Large law firms are using a tool made by OpenAI to research and write legal documents. What could go wrong? Tweets: @robertjbateman , @carnage4life , @stokel , and @gteninbaum Tweets: Robert Bateman / @robertjbateman : I am quoted in this excellent Wired piece about the use of large language models by law firms. I think the elephant in the room is whether data used to train LLMs has been obtained in a GDPR-compliant way. And if not, whether they can be used lawfully. https://www.wired.com/... Dare Obasanjo / @carnage4life : The UK's second largest law firm by revenue has gone all in on using OpenAI to answer legal questions and create briefs. Things I've predicted about this space I expected to happen in months or years are happening in days. This is a tech revolution. https://www.wired.com/... @stokel : Generative AI's new use case? Legal documents. My latest for @WIRED looks at one company utilising it, and the risks involved https://www.wired.com/... Gabe Teninbaum / @gteninbaum : Not sure I agree with all the conclusions but the fact Wired is covering law/tech overlap is notable! https://twitter.com/...
Context & Ripple Effects
When Wired sat down with David Wakeling of one of the world's largest law firms, Allen & Overy was among the first majors to put OpenAI-built Harvey into production for legal research and drafting — an early proof point that generative AI could handle work long done by junior associates. The piece also surfaced the sharpest open question, raised by Robert Bateman: whether the data used to train these models was obtained in a GDPR-compliant way at all.
The arc since has validated the bet: Harvey moved from pilot curiosity to talks to raise $600M at a $2B+ valuation while weighing an acquisition of legal-research firm vLex, and adoption spread beyond firms to US judges themselves.
First-order effects
- Allen & Overy gains a working template for AI-assisted legal drafting while carrying unresolved GDPR exposure on its training-data provenance — the same compliance question Bateman flags applies to every client deliverable produced with the tool.
Second-order effects
- Competing large firms must either adopt equivalent tools or justify premium billing for work rivals produce faster, pressuring the billable-hour model that generative AI is already reshaping across professional services; Harvey's scale-up and possible vLex purchase would consolidate the legal-AI supply side those firms now depend on.
Third-order effects
- If courts extend adoption — as coverage of US judges testing generative AI suggests — drafting and review become shared human-AI workflows across the entire litigation chain, pushing regulators toward explicit rules on training-data legality and model accountability in professional practice.
The trend: Legal work is moving from firm-owned associate labor to rented foundation-model tooling, with compliance regimes like GDPR and court-side adoption setting how fast the shift can go.