Cleary Gottlieb
piloting Gemini Enterprise for Legal with contract-review and regulatory-scanning skills and pre-built agents
Accueil › AI Use-Case Library › AI in legal work: 6 real deployments
AI in legal work: 6 real deployments
Contract review, drafting, regulatory rewriting and the expert reports that went wrong.
6deployments
5in production or with results
3with a reported outcome
0halted or reversed
Aug 30, 2026last updated
Law firms and agencies are putting models on documents. Alongside the platform deployments, the entries record the drafts submitted with model artefacts still in the text and the rules rewritten by a tool defaulting to rescission.
Legal 4 deployments
Harvey
Deployed Harvey II legal AI platform featuring Memory that learns individual lawyer drafting style across Harvey, Word and Outlook, and Tenet, an in-house proprietary legal model trained on mock disputes and case files using a version of Moonshot's Kimi K3
Knighthawk Engineering
Expert witness Josh Autenrieth used ChatGPT to produce approximately 85-90% of an expert report defending 3M in the Watson Grinding fatal explosion lawsuit, prompting it to show 3M is at zero fault
Reported: Jury awarded plaintiffs $61M assigning 30% fault to 3M; 350 pages of ChatGPT prompts entered discovery and were used to impeach the expert witness at trial
Thomson Reuters
Thomson Reuters is deploying AI across legal, tax and regulatory workflows while reshaping its engineering organization.
Reported: The company confirmed up to 500 engineering layoffs and plans to hire more than 250 net-new, mostly senior AI-native engineers over two years.
Government & Public Sector 2 deployments
Rep. Anna Paulina Luna
Used Claude to draft a June congressional amendment, submitting text that still contained the phrase 'Claude responded'
US Department of Housing and Urban Development
HUD uses the DOGE-built SweetREX AI tool to analyze and rewrite federal housing regulations under a default assumption that rules should be rescinded.
Reported: DOGE claims a 90% reduction in rulemaking time and says the tool made decisions on 1,083 regulatory sections.
Every entry names the organisation and links its source. Outcome figures are quoted as reported, never estimated. Vendor announcements without a named customer are excluded. Halted and reversed deployments are kept on purpose.