Law Firms Say Anthropic Ties Privacy Terms to Minimum Spend

Bloomberg Law reported on 22 October 2026 that law-firm innovation leaders came away from a private meeting with Anthropic in August having heard that the company would not let them negotiate key contract terms around privacy and data security unless they committed to a minimum amount of spending. Bloomberg Law attributes that account to three sources who attended the meeting.

The report sets that against a survey published in September by the International Legal Technology Association (ILTA): 44% of firms said they were using or exploring Claude, but among those firms, just 7% said they had fully rolled it out to attorneys, Bloomberg Law reported.

Business Engineer · ENTERPRISE AI ADOPTION

A short Business Engineer video on the state of enterprise AI adoption. It makes no claim about Anthropic, the law firms or the survey in this piece.

What Was Reported

The meeting was billed as a “fireside chat” with Anthropic’s new head of Claude for Legal, Robert Mahari, at a convention in Nashville, Tennessee, according to Bloomberg Law. Mahari was a few weeks into the role at the time. He has a JD-PhD from Harvard and MIT and founded Akiva AI before joining Anthropic, the report says.

Beyond the contract terms, attendees “also left with the impression they’re more or less on their own for training,” Bloomberg Law reported. Its account of how the presentation was received draws on interviews with five people who attended and were granted anonymity to describe a closed-door discussion.

“They need to wake up,” one of the people in the room said of Anthropic, according to Bloomberg Law. “They do need to hear that we’re not impressed.” Another attendee told Bloomberg Law the presentation “landed awkwardly in a room full of people who are trying to see if they can actually deploy Claude and are under pressure from their stakeholders.”

One attendee said Anthropic is not far off the pace on the privacy terms in its base contracts, but that his firm often negotiates privacy terms and Anthropic isn’t willing to do that, Bloomberg Law reported. Another attendee objected to Anthropic’s decision not to offer zero data retention for its Fable 5 model, the report says. Bloomberg Law notes that it is itself a legal technology provider.

What Was Posted on X

Choblin’s post relays Bloomberg Law’s report: Anthropic’s head of Claude for Legal told law-firm leaders that key privacy and data security terms are not negotiable without a minimum spend, plus the survey’s 44% exploring and 7% fully rolled out. It attaches screenshots of the article. As we read it, the post’s last line, that some firms like Claude but are less impressed with Anthropic, matches the split in Bloomberg Law’s interviews.

The Numbers: Interest Versus Rollout

ILTA’s 2026 technology survey spanned more than 500 law firms, all ILTA members, from a dozen countries and representing 140,000 lawyers, according to LawSites’ 14 September write-up of the results. In it, 94% of firms said they were using or exploring generative AI tools. Asked which tools, firms put Microsoft 365 Copilot first at 76%, with Claude and Thomson Reuters CoCounsel at 44% each and Harvey at 43%.

The rollout figure is the gap. Among the firms using or exploring Claude, just 7% said they had fully rolled it out to attorneys, according to Bloomberg Law’s reading of the same survey. Taken across all surveyed firms, that is roughly 3 in 100 on our arithmetic (7% of 44%). LawSites reported that only two products had been fully deployed by at least half of respondent firms’ lawyers: Microsoft Copilot, which the survey lists as the version formerly called Bing Chat, at 52% and Westlaw Advantage at 50%.

Testing is a different picture. In LawSites’ list of products that firms are currently testing or piloting, Claude was third at 68%, behind two iManage products.

What Anthropic Says

Anthropic declined to make Mahari available for an interview, and he didn’t respond to a request for comment, Bloomberg Law reported. An Anthropic spokesperson said in an email that the company doesn’t discuss “commercial terms” and pointed to two support pages.

According to Bloomberg Law’s account of those pages, Anthropic doesn’t train its models on enterprise customers’ data by default, and customers can select which services they connect to Claude. Anthropic offers zero data retention, meaning prompts aren’t stored, but retains inputs and outputs of its Mythos class models for 30 days for “safety monitoring”, the report says, citing the legal support page. “We believe Claude’s commercial offerings can be configured by lawyers in a way that maintains privilege and confidentiality,” the company says on its Claude support page, as quoted by Bloomberg Law.

What Firms Say They Need

Not every firm objects. Ogletree Deakins has been running a pilot with Claude since February, and its chief data and AI officer, Timothy Fox, told Bloomberg Law: “Our deployment of Claude has been one of our most successful.” Fox also said “it is not a product that is geared for legal, and so they don’t design their enterprise security like a legal platform would.”

Training is where firms compare Anthropic with legal vendors. Legora hosted multiple full-day training sessions with attorneys at Crowell & Moring, and Polsinelli hosted more than 90 trainings with Thomson Reuters to get up to speed on CoCounsel, Bloomberg Law reported. One attendee said that without Anthropic’s support, his firm might have to pay consultants to train its attorneys. Alex Finkel, chief innovation officer at Offit Kurman, said Anthropic is still building that infrastructure: “They are at the very beginning of that journey.”

Bloomberg Law also reported that Google announced a suite of products for lawyers the day before the Nashville meeting, and that OpenAI made a similar announcement a few weeks later. “We no longer feel the pressure because OpenAI closed the gap, if not surpassed Claude in many ways,” one innovation leader said.

One meeting attendee summed up the takeaway: “there was a strong sense that we as an industry probably don’t rate significantly high on Anthropic’s universe of expected revenue, and therefore the level of attention that we would get as an industry is relatively low.” Mandeep Singh, global head of technology research at Bloomberg Intelligence, told Bloomberg Law: “They look for new markets given they do have pressure on sustaining that AI growth.”

The Structural Read

As we read it, the minimum spend turns a contract term into a price: according to the attendees Bloomberg Law interviewed, a firm that wants to negotiate privacy and data security terms first commits to a level of spending.

As we read it, legal vendors sell the deployment as well as the software. Legora ran full-day training sessions at Crowell & Moring and Polsinelli hosted more than 90 trainings with Thomson Reuters, Bloomberg Law reported, while attendees left Anthropic’s meeting feeling “more or less on their own for training.”

As we read it, the attendees’ own explanation is about attention rather than the model: one said the legal industry probably doesn’t “rate significantly high on Anthropic’s universe of expected revenue.”

Three Implications

FOR LAW FIRMS One meeting attendee told Bloomberg Law that without Anthropic’s support, his firm might have to pay consultants to train its attorneys.

FOR LEGAL AI VENDORS As we read it, training and legal-specific security are where legal vendors differ from Claude in the accounts firms gave Bloomberg Law; Ogletree Deakins’ Timothy Fox said legal vendors “know what a law firm is looking for.”

FOR ANTHROPIC Bloomberg Law reports that Google and OpenAI have both announced legal products and hired former legal tech leaders in roles that appear to be similar to Mahari’s.

The Business Engineer Lens

This story maps onto the Business Engineer framework Enterprise AI: From Software to Substrate.

The framework argues that in enterprise software “The value proposition shifts from features to integration depth.”

As we read it, the law-firm account is about that integration step: firms report trying Claude widely but rolling it out rarely, and say the training and negotiated terms that would embed it are what they are missing. The framework was published in December 2025 and does not discuss Anthropic’s legal sales or this survey.

What Is Not Established

Bloomberg Law’s report does not give the size of the minimum spend, and Anthropic has not confirmed or described the condition; its spokesperson said the company doesn’t discuss commercial terms. The account of the meeting comes from attendees speaking anonymously. We read ILTA’s results through LawSites and Bloomberg Law, not the full survey. The 7% figure is Bloomberg Law’s, and the roughly 3 in 100 is our arithmetic.

Connecting the Coverage

On 9 September we covered Harvey’s $550 million round at a $15.5 billion valuation in our piece on Harvey’s round, and on 21 October we covered Legora’s $550 million Series E at an $8.8 billion valuation, with annual recurring revenue of “more than $200 million”, in our piece on Legora’s round. Both are legal AI vendors in the table above. What’s new today, as we read it, is a first account of how Anthropic sells Claude to law firms and how those firms say they received it.

The Bottom Line

As we read it, the reported gap is between trying Claude and deploying it: 44% of surveyed firms are using or exploring it, and 7% of those have rolled it out fully. Attendees told Bloomberg Law that negotiating privacy terms requires a minimum spend, and that training is largely left to the firms.