Legal Tech Rewind: June 2026, the Month a Frontier AI Model Got Switched Off

Legal Tech Rewind is our month-by-month look back at 2026 for in-house legal teams: what happened, why it mattered, and what it looks like with the benefit of hindsight. Published October 2026.

June in one sentence

A government order showed that the AI under your legal tools can disappear overnight, courts started disagreeing about whether AI chats are discoverable, and the agent products kept moving deeper into in-house workflows.

1. A frontier AI model was switched off for everyone

On June 9, Anthropic released Claude Fable 5, with the related Mythos 5 in limited release. On June 12, the US Commerce Department sent a letter requiring an export license before any foreign national, inside or outside the US, could access either model. Anthropic couldn't verify every user's nationality in real time, so it disabled both models for all customers, including through AWS, Google Cloud and Microsoft's cloud. Commerce lifted the controls on June 30, and Anthropic restored access on July 1. CSIS described it as the first use of this export-control authority over a commercially deployed frontier model.

Why it mattered: almost every legal AI product runs on a handful of frontier models, often from a single provider. This was the first time many legal teams saw what happens when one becomes unavailable with no warning.

Since then: the episode resolved in under three weeks, but the lesson stuck. Your vendor contracts should say what happens if the underlying model is withdrawn: can the vendor switch models, how much notice do you get, and does the quality of the work you rely on change? (We turned this into a checklist in our agent governance post.)

2. Courts split on whether AI chats are protected

In February, United States v. Heppner in the Southern District of New York held that a defendant's chats with a consumer AI tool weren't privileged or protected work product. In June, two state courts went the other way:

• In Tate Group Automotive v. Legacy Automotive Capital, the Texas Business Court held that a party principal's ChatGPT conversations were protected work product, though it still required the party to identify which discovery documents had been uploaded.
• In Assini v. Hayward, a New York trial court quashed subpoenas for a pro se litigant's AI prompts and outputs as trial-preparation material.

Why it mattered: whether your executives' AI chats can be discovered now depends on the court and the facts, including the tool's confidentiality terms and whether counsel directed the work.

What to do: steer anything litigation-related into enterprise AI tools used at counsel's direction, and say so in your AI policy.

3. AI-citation sanctions graduated to suspensions

In the Northern District of Mississippi, Judge Sharion Aycock suspended two lawyers from practicing in the district for two years and fined them after AI-assisted filings contained fabricated citations. Two local counsel were fined and removed from the case. In New York (reported July 2), the Appellate Division sanctioned a lawyer and his firm $10,500 combined for an appellate brief citing nonexistent cases. The lawyer had violated his own firm's AI verification policy. Florida's new citation rule also took effect on June 15.

Why it mattered: when your outside counsel gets suspended mid-case, it's your matter that's disrupted.

4. The agents moved further into in-house workflows

• Thomson Reuters opened early access to next-gen CoCounsel Legal on June 22. It turns a matter description into one plan and one piece of work grounded in Westlaw and Practical Law, with matter Workspaces and integrations including iManage, NetDocuments, Icertis and Microsoft 365.
• Harvey became available as an agent inside Microsoft 365 Copilot and Copilot Cowork on June 16, and added 500 in-house contracting tasks to its Legal Agent Benchmark.
• Perplexity entered legal with "Computer for Counsel," relying on connectors (including Midpage for case law) rather than building its own legal database.
• Spellbook launched Autonomous Contract Management for in-house teams, covering intake, review and post-signature insight. Artificial Lawyer called it a "CLM killer," though Spellbook's CEO says it isn't a CLM.

Why it mattered: legal AI started reaching business users directly inside Microsoft 365, which means access control is now a legal ops problem, not just an IT one.

5. Investors bet on in-house first

Wordsmith raised a $70 million Series B for an AI "front door" for legal requests. Sandstone raised a $30 million Series A, led by Lightspeed, for what it calls legal relationship management. On the M&A side, Relativity bought Word-based drafting company Gavel, and Clio bought Canadian case law company Jurisage.

Why it mattered: more vendors are now building specifically for corporate legal intake and triage, not adapting law firm tools. More choice means better pricing leverage.

6. Deloitte put numbers on the in-house future

A Deloitte Legal report projected that AI agents could handle 30% of corporate legal work within three to five years. Respondents expected hourly-billed work to fall from 72% to 44% within two to three years. The striking number: 58% of GCs said their outside firms rarely or never raise AI benefits proactively, and only 4% had directly seen AI benefits from their firms' AI use. These are projections and the sample size wasn't disclosed, so use them as conversation starters, not benchmarks.

Since then: Thomson Reuters' July research found the same gap (70% of corporate respondents expect AI productivity gains from their firms, but only 6% say most firms deliver). The pricing conversation with your firms is overdue.

7. Regulators kept delaying

The European Parliament adopted the AI Act delay on June 16 (423 votes for, 57 against, 174 abstentions), pushing high-risk obligations to December 2027 and August 2028. Colorado's original June 30 AI Act date passed quietly, because the law had already been replaced in May. New York's legislature passed a package of AI transparency bills that Gov. Hochul has until year-end to act on.

What to do with June's lessons now

• Add model continuity to vendor contracts. Ask for notice of model changes, substitution rights and service credits for model outages. Start with your generative AI legal assistant vendors.
• Update your AI policy for discovery. Litigation-related AI use should happen in enterprise tools, at counsel's direction, and be preserved under your legal hold process.
• Raise AI pricing with your firms. Use the Deloitte and Thomson Reuters numbers, and track results through e-billing and spend management.
• Look at in-house-first intake tools. The legal intake, workflow and service delivery market got a lot more competitive this year.

June's lesson: you don't just have a vendor. You have a vendor's vendor's model, and someone should own that risk.

More from the 2026 Legal Tech Rewind: ← May 2026 · July 2026 → · The full 2026 Rewind: 9 lessons

Sources

• CSIS, "The Department of Commerce Restricted Access to Anthropic's Latest Models. What Comes Next?" (June 16, 2026): csis.org
• IAPP, "US government order forces commercial suspension of two frontier AI models" (June 15, 2026): iapp.org
• Anthropic, "Redeploying Claude Fable 5" (June 30, 2026): anthropic.com
• Al Jazeera, "US lifts restrictions on Anthropic's powerful AI models Fable and Mythos" (July 1, 2026): aljazeera.com
• Spencer Fane, "Courts Begin Shielding AI Prompts and Outputs From Discovery" (July 28, 2026): spencerfane.com
• Harvard Law School Forum on Corporate Governance, "Are AI Legal Chats by Non-Lawyer Officers and Directors Discoverable?" (July 22, 2026): corpgov.law.harvard.edu
• JD Journal, Mississippi judge disqualifies lawyers over AI misuse (June 9, 2026): jdjournal.com
• Minnesota Lawyer, NY court sanctions lawyer for AI fake citations (July 2, 2026): minnlawyer.com
• LawSites, Thomson Reuters opens early access to next-gen CoCounsel Legal (June 22, 2026): lawnext.com
• Harvey, Copilot Cowork launch (June 16, 2026): harvey.ai
• Harvey, Legal Agent Benchmark in-house contracting (June 12, 2026): harvey.ai
• LawSites, Perplexity's Computer for Counsel (June 25, 2026): lawnext.com
• Artificial Lawyer, "Spellbook Launches 'CLM Killer' ACM" (June 30, 2026): artificiallawyer.com
• Artificial Lawyer, "Wordsmith Raises $70m Series B" (June 3, 2026): artificiallawyer.com
• Artificial Lawyer, "Sandstone Raises $30m For AI-Native Inhouse Teams" (June 9, 2026): artificiallawyer.com
• LawSites, Relativity acquires Gavel (June 15, 2026): lawnext.com
• LawSites, Clio acquires Jurisage (June 10, 2026): lawnext.com
• Artificial Lawyer, "AI Agents Will Handle 30% of Inhouse Work: Deloitte" (June 30, 2026): artificiallawyer.com
• Thomson Reuters, Future of Professionals 2026, law firm findings (July 21, 2026): thomsonreuters.com
• Sidley Data Matters, EU lawmakers and the AI Act delay (June 22, 2026): datamatters.sidley.com
• Transparency Coalition, AI Legislative Update (June 26, 2026): transparencycoalition.ai

This article is general information, not legal advice.