Anthropic Courts the Enterprise, and the Data Governance Questions Come With It

Anthropic is no longer content to be the model that researchers admire and procurement teams ignore. A recent IDC report, titled "The Transformation of Anthropic (and What to Do About It)," argues that the company has spent the first half of 2026 building the scaffolding of an enterprise vendor, and that buyers should revisit their large language model and agent evaluations as a result. The consultancy's framing is blunt: no frontier model company is mature enough today to be judged as a standalone enterprise AI provider, but Anthropic is "running at full speed to get there before its competitors."

For anyone who reads technology news through the lens of data protection and user rights, the more interesting story sits underneath the adoption numbers. When a model vendor moves from experimental tooling into core enterprise systems, it stops being a curiosity and becomes a data processor. That shift carries real legal weight under the GDPR in Europe and the California Consumer Privacy Act in the United States, and it changes the questions a CISO is obligated to ask.

What IDC actually reported

IDC has tracked Anthropic's moves over roughly six months and counted more than 100 public interactions between January and May 2026: official announcements, release notes, blog posts, partner deals, hiring news, and policy moves. The consultancy reads these as deliberate steps to expand distribution, deepen enterprise support, target specific industries, and, in its own phrasing, enhance "stickiness," which is a polite term for lock-in.

The usage data shows why the effort matters. Per IDC's FERS Survey from March 2026, only 19 percent of enterprises use Claude extensively and 25 percent are actively evaluating it. OpenAI and Google sit well ahead, represented in roughly 42 percent and 38 percent of organizations respectively. Yet Anthropic's revenue mix is heavily commercial. Citing The Information, IDC notes that about 86 percent of Anthropic's projected 2025 revenue came from enterprise sales, compared with around 40 percent for OpenAI. In raw dollars OpenAI's business revenue was larger, $5.2 billion against Anthropic's $3.9 billion as of January, but the proportional dependence on enterprise customers tells you where Anthropic has placed its bet.

That bet became visible in November 2025, when Anthropic began moving enterprise customers away from seat-based pricing toward usage-based pricing, and it has continued through initiatives like the Claude Partner Network.

Why the legal basis matters more as Claude scales

A pricing change and a partner network sound like commercial housekeeping. For a data protection officer, they are the start of a due diligence checklist. The moment a Claude deployment touches customer records, employee data, or anything that qualifies as personal data, the deploying company becomes a data controller and the model provider becomes a processor. Under Article 28 of the GDPR, that relationship has to be governed by a written contract that specifies what data is processed, for how long, under what security measures, and whether sub-processors are involved.

Usage-based pricing complicates that picture in a subtle way. Seat-based licensing tends to keep usage predictable and bounded. Usage-based models reward sending more data through the system, which is exactly the behavior privacy regulators look at when they assess data minimization, one of the GDPR's core principles. The CCPA, and its successor framework the California Privacy Rights Act, adds its own demands around purpose limitation and the right of consumers to know what is collected and to opt out of certain processing. An agentic system that autonomously pulls from internal databases and external tools can blur the line on purpose limitation faster than a human-supervised workflow ever did.

The agent problem for users and companies

IDC frames the opportunity in terms of "a multi-LLM or an agentic AI strategy," and that word, agentic, is where the user impact concentrates. A chatbot answers a question and forgets it. An agent takes actions: it reads a ticket, queries a CRM, drafts an email, updates a record. Each of those steps can involve personal data, and each step is a potential disclosure point if the agent's permissions are scoped too broadly or its logs retain more than they should.

For companies, the compliance implication is that traditional vendor risk assessments, built for static software, do not map cleanly onto systems that make decisions. Auditing an agent means understanding not just what data goes in, but what the model can do with it, what it stores, and whether its outputs can leak information from one customer's context into another's. Data residency adds another layer, since GDPR transfers of personal data outside the European Economic Area require a valid legal mechanism such as Standard Contractual Clauses.

For users, the impact is harder to see and therefore easier to overlook. When an enterprise routes support requests or HR queries through an AI system, the people on the other end rarely get told which model handled their data or where it ran. The right to be informed, a foundation of both GDPR and CCPA, depends on the deploying company passing transparency obligations downstream. A vendor racing for enterprise share has every incentive to make integration frictionless, and frictionless integration is often the enemy of clear disclosure.

What changes

IDC's advice to CIOs and CISOs is to reassess where Claude fits. The honest version of that advice includes the parts the report does not dwell on. Reassessment should mean pulling the data processing agreement and checking the sub-processor list, confirming retention and training-data policies in writing, and verifying that any agentic deployment respects the same access controls a human employee would face. It means treating model output logs as the sensitive records they often are.

The enterprise ecosystem's preference for a vendor-neutral, multi-LLM approach actually helps here, because it forces a level of abstraction that makes data flows easier to audit and harder to lock in. Lock-in, the "stickiness" IDC praises as a business virtue, is the same property that makes it expensive to walk away from a vendor whose privacy practices no longer satisfy a regulator.

Anthropic has built a reputation partly on safety research, and that gives it a credible story to tell enterprise buyers worried about governance. The test now is whether the contractual and technical guarantees keep pace with the sales push. A model becomes an enterprise provider not when it appears in more IT conversations, but when it can satisfy a data protection audit without the buyer having to take anything on faith.