Your Briefing: What's Happening in Legal AI Right Now
Your July round-up of the developments reshaping legal practice
From corporate general counsel overtaking law firms in AI adoption to the Kirkland-Palantir announcement still provoking conversation in the market a month later, to a Chinese open-weight model raising both cost and security questions, to a law school laptop ban. Here's everything you need to know from the last month.
By the Numbers
• 83% of US legal practitioners now use AI at work, per Bloomberg Law's first State of Practice survey (2026)
• 1,000+ Kirkland & Ellis lawyers set to use its new Palantir-built AI platform for private equity fundraising following $500 million technology investment announcement
• 70% of general counsel plan to invest in new AI technology within the next 12 months (FTI Consulting / Relativity, 2026)
• 60-90% cheaper: how much less it costs to run Chinese open-weight models than leading US systems (CNBC, 2026)
Top Stories
Adoption
In-house counsel is setting the pace on AI
The adoption story has flipped. Law firm leadership was early to champion AI, but general counsel are now moving faster. FTI Consulting and Relativity's latest General Counsel Report finds 87% of legal departments now use generative AI, nearly double last year's 44%, with summarization, clause identification, and transcription the leading use cases. A separate survey of 121 legal leaders found GCs increasingly using AI to cut costs, insource work once sent to outside counsel, and push back on law firm pricing.
Firms with a formal technology roadmap jumped to 53% of legal departments, up from 25% a year ago, a sign that in-house AI use is shifting from experimentation to strategy. For outside counsel, that shift is starting to be felt directly in panel reviews and instruction decisions.
→ FTI Consulting: The General Counsel Report 2026
→ Law.com: Law Firms, Clients and AI in 2026
Business of Law
Firms are racing to build proprietary AI tools. The payoff still hasn't been shown
Big Law is increasingly building proprietary AI tools rather than relying solely on off-the-shelf products, and several firms are now weighing whether to license those tools to clients, a move that could open subscription revenue but also risks accelerating client insourcing. A separate report on in-house counsel across Asia found firms investing millions in AI without it translating into lower bills, clearer pricing, or measurable value, what the piece calls the "AI value gap."
Worth noting what's missing from the coverage of these proprietary builds: strategic rationale, revenue potential, client stickiness, differentiation, but no published data from any firm showing a homebuilt system has actually reduced costs or improved outcomes for a client yet. That's not the same as vendor-built AI, where efficiency and adoption numbers are already out there and measurable. Until a firm can point to results rather than intent, the case for a multi-year internal build remains a bet on a promise, not a demonstrated return.
→ Law.com: Law Firms Are Building Their Own AI Tools. Should They Share Them?
→ Law.com International: The AI Value Gap
BigLaw
The Kirkland-Palantir deal keeps making waves a month later
Kirkland & Ellis and Palantir announced a multi-year partnership on June 4 to build a proprietary, Palantir AIP-based platform for private equity fund formation, covering documentation, side letters, investor tracking, and compliance, with more than 1,000 Kirkland lawyers expected to use it. It's part of Kirkland's broader $500 million infrastructure bet, backed by an AI team of roughly 180 engineers and data scientists.
The story kept developing into July. A July 22 Legaltech News analysis framed the deal as a bellwether: elite firms with the balance sheet to build custom AI are pulling further ahead, while everyone else now has to choose between building, licensing, or buying a vertical platform of their own.
→ Kirkland & Ellis: Kirkland & Ellis and Palantir Partner to Transform PE Fundraising
→ Law.com Legaltech News: The AI Arms Race in Private Equity
Tools Worth Watching
China's open-weight models just became Washington's problem, and law firms need to pay attention
Moonshot AI released Kimi K3 in late July, a 2.8-trillion-parameter open-weight model downloaded more than 930,000 times in its first week, with demand so heavy Moonshot briefly paused new subscriptions. The performance-parity claim is real but nuanced: on the Artificial Analysis Intelligence Index v4.1, Kimi K3 scored 57.1, behind Claude Fable 5 and GPT-5.6 Sol but ahead of Claude Opus 4.8, and it ranked first on the Frontend Code Arena, ahead of every closed competitor tested. It also posts the strongest GPQA Diamond score of any open-weight model to date. It's not uniformly the best model available, but it's close enough on enough benchmarks that "good enough, and open" is now a credible pitch to firms and legal vendors.
The cost case is the other half of the pitch. Chinese open-weight models like DeepSeek, Qwen, and GLM now run 60-90% more cheaply than the leading Anthropic and OpenAI systems, and Chinese-origin models have accounted for as much as 46% of weekly US enterprise token traffic this year. Legaltech News covered the legal-industry angle on July 27, noting vendors are starting to evaluate these models for cost-sensitive, high-volume work.
Security is where it gets complicated, on two levels. At the firm level, the risk is entirely about deployment mode, not the model itself: self-hosting the open weights keeps prompts off Chinese servers entirely, but the hosted, consumer-facing versions route data through China under Chinese law, with no BAA, DPA, or enterprise data-residency option, which is why regulated industries are being told to avoid the hosted service for confidential or client data. Self-hosting isn't trivial either; it requires real GPU infrastructure and dedicated staff.
At the national level, Booz Allen's June 5 report tested four Chinese frontier models and found that three of four produced significantly more vulnerable, obfuscated code when prompted with a US government persona. Combined with reports that the Trump administration is fighting internally over how to counter Chinese open-source AI's rapid US adoption, that Treasury is examining Chinese AI firms for IP theft, and that lawmakers have opened probes into US companies' use of these models, firms adopting them are now navigating both a client-confidentiality question and a geopolitical one.
→ CNBC: Chinese AI Models Are Gaining Ground With US Companies
→ Law.com Legaltech News: What Chinese Open-Weight AI Models Mean for Legal Tech
→ Axios: The Secret Trump Administration Battle to Fight Chinese AI
Education
UChicago Law bans laptops from 1L classrooms as part of a sweeping new AI strategy
The University of Chicago Law School released its formal AI strategy statement on July 9, "Rethinking Legal Education in the AI Era," and it's more restrictive than most. Starting in the 2026-2027 academic year, all nine core 1L courses will pilot a coordinated no-device policy, with in-class exams administered without internet, electronic files, or apps.
The philosophy: writing without AI comes first, as the foundation, before AI is layered on for research, revision, and oral argument prep. It's a notable counterpoint to the more permissive approaches schools like Stanford and Washington University have taken.
→ LawSites: UChicago Law Bans Laptops from 1L Classrooms
Governance
Connecticut passes a sweeping AI law, and malpractice insurers are taking notice
Connecticut signed the AI Responsibility and Transparency Act (the "CART Act") into law on June 2, one of the most comprehensive state AI statutes to date, covering employment-related automated decision tools, consumer chatbots, generative-AI provenance, and platforms used by minors. The first provisions phase in starting July 1, with most obligations effective October 1, a narrow window for firms serving Connecticut clients to get ahead of it.
Meanwhile, on the risk side, more than half of the major insurers covering 80% of Am Law 200 firms have seen a rise in AI-related malpractice claims, and Verisk is rolling out new generative-AI exclusions that could reshape coverage market-wide. AI governance is no longer just an ethics question. It's now an underwriting one.
→ Holland & Knight: Connecticut Enacts Comprehensive AI Legislation
→ Law.com: Rising AI Mistakes in Legal Pose Quandary for Law Firms' Insurance Policies
