Despite hallucinations emerging in legal filings, Big Law intensifies its AI commitment with Anthropic’s latest release

(SeaPRwire) –   Judges have imposed sanctions, bar associations have issued warnings, and across U.S. courtrooms, attorneys have been caught submitting legal briefs citing nonexistent cases—fabricated precedents generated by AI tools that present false information with the appearance of credibility.

This growing concern has not deterred major law firms from embracing artificial intelligence. On the contrary, the industry appears to be accelerating its adoption.

On Tuesday, Anthropic launched its most comprehensive integration effort to date for legal workflows, introducing more than 20 new connections with widely used legal technology platforms. The rollout includes 12 specialized AI plugins tailored to specific legal roles, spanning areas such as mergers and acquisitions due diligence and employment handbook creation. Additionally, a cross-platform Microsoft 365 integration embeds Claude as a unified contextual agent across Word, Outlook, Excel, and PowerPoint. Underlying this update is the Claude Opus 4.7 model, which achieved a score of 90.9% on Harvey’s BigLaw Bench—the legal sector’s premier benchmark that rigorously tests large language models against real-world legal tasks, aiming ultimately to replace traditional billable work hours.

Anthropic highlighted several high-profile law firm partners during the announcement: Freshfields, Quinn Emanuel Urquhart & Sullivan, Holland & Knight, and Crosby Legal are all actively using Claude in ongoing legal matters. These firms jointly disclosed their collaboration with Anthropic, along with the fact that numerous legal AI platforms—including Harvey, Legora, Solve Intelligence, and Eve—are built on top of Claude’s core models. More than 20,000 legal professionals registered for Anthropic’s latest legal-focused webinar, marking the largest such session the company has ever hosted.

Gerrit Beckhaus, Partner and Co-Head at Freshfields, stated that Claude has become an “essential component” of the firm’s proprietary AI solutions. He added that the firm is collaborating with Anthropic to co-develop agentic workflows capable of managing complex, multi-step legal processes from start to finish.

Christopher D. Kercher, Partner and founder of Quinn Emanuel’s AI & Data Analytics practice, explained that he developed the firm’s litigation platform entirely around Claude despite having no formal coding background. The motivation came from a real trial requirement. “The breakthrough was treating Claude as if it were a senior attorney joining mid-case,” Kercher said. “We onboarded it with chronologies, key excerpts, and thematic summaries just as we would a new partner. The resulting work product far exceeds anything I could have produced alone—or even imagined achieving.”

This marks a sharp contrast to just weeks earlier, when Sullivan & Cromwell—another elite white-shoe firm—was exposed for including a fabricated case citation in a bankruptcy court filing. In response, partner Andrew Dietderich apologized directly to the judge, writing, “We deeply regret that this has occurred.”

Addressing the Hallucination Challenge?

Anthropic’s solution to the problem of AI-generated inaccuracies is called “grounding,” a technical approach that ensures Claude only references live, verified sources—such as Westlaw’s case law database, CourtListener’s repository of authentic judicial opinions, or iManage document systems—rather than relying on internal memory or synthetic content generation. According to Anthropic, an AI that consults actual legal documents behaves fundamentally differently from one that fabricates responses based solely on training data.

“In litigation, a plausible-sounding but incorrect statement can cause more damage than no answer at all,” said Jay Madheswaran, CEO and co-founder of Eve, a legal AI company built on Claude. His organization evaluates every model against “more than 24 specialized legal criteria—including citation accuracy, prevention of unverified case quotes, memory leakage, and proper refusal protocols.” Madheswaran emphasized that Claude consistently outperforms other models in internal testing, particularly in grounding and faithful citation practices. “That’s why the most critical stages of our workflow rely exclusively on Anthropic’s technology.”

Jake Lauritzen, CTO of Legora, noted that Claude Opus 4.7 demonstrates “improved consistency when processing lengthy documents, better responsiveness to nuanced instructions, and enhanced reliability in high-stakes legal environments” compared to previous versions.

The business implications for Anthropic are substantial. The company revealed that legal professionals now constitute the largest user base on its Cowork platform. Tuesday’s announcement positions Anthropic not merely as a behind-the-scenes model provider—like the engine powering platforms such as Harvey or Legora—but as an active participant in direct legal workflows. This places the company in a potentially conflicting relationship with established industry players; for instance, Thomson Reuters both supplies Claude access to its Westlaw legal database and markets competing AI products of its own.

Beyond serving BigLaw, Anthropic is advancing an access-to-justice narrative. Approximately 80% of civil litigants appear in court without legal representation, according to the company. Through partnerships with organizations such as Free Law Project and Courtroom5—whose connectors will be available to Claude users at no extra charge—Anthropic frames its expansion not simply as a productivity tool for hourly-billing firms, but as a potential democratizing force.

“Most people don’t realize they have legal rights until those rights are too difficult to exercise,” said Sonja Ebron, CEO and co-founder of Courtroom5. “Claude can now meet them exactly where they are—in the moment they’re frightened and urgently seeking answers.”

The legal profession has made its strategic choice. Whether these technological safeguards prove sufficient will ultimately be determined through judicial review.