A note before I start: though the story below comes from my work, I’m writing as myself. I’m not a spokesperson for any organization, and which enterprise platform my employer adopts — if any — is a decision that belongs to senior leadership, not to me.
On February 11, a clinician asked a simple question at an AI committee meeting: can we get a small Enterprise plan with a HIPAA Business Associate Agreement from one of the major AI vendors? The 300+ person healthcare specialty group has real use cases that hypothetically could touch patient data — clinical workflow tools, text extraction from reports, scheduling models trained on productivity history. None of that is safe to try on a consumer account.
I went to work that afternoon. Two months and change later, two of the three major vendors have responded cleanly enough to compare quotes. The third has gone completely dark.
Why we need Enterprise
If an AI tool is going to touch protected health information — or even get close enough that a user might paste something into it by mistake — the vendor has to sign a Business Associate Agreement. The BAA is what pulls them under the HIPAA umbrella with us. Without it, sending a record into a chatbot is the same as emailing it to a stranger. It’s a breach the day you hit submit, not the day someone finds out.
There’s a second reason Enterprise matters, and it’s the one that should matter most to any patient reading this. On consumer and even most business-tier plans, vendors can use your prompts, uploaded files, and model outputs to train future versions of their models. Enterprise plans flip that default — customer data is contractually excluded from model training. Nothing about a patient’s visit belongs in anyone’s training corpus, and Enterprise is how we make that contractual rather than aspirational.
Free and personal accounts don’t get either of those protections. Team plans usually don’t either. Every vendor I contacted routes both the BAA and the training-exclusion commitment through the Enterprise tier. So “can we get a small Enterprise plan” isn’t optional paperwork — it’s the floor for any serious AI work inside a healthcare organization.
What we already have
Our users already have Microsoft Copilot through our Microsoft 365 Enterprise Agreement, and Microsoft’s BAA covers it. For a chatbot — summarizing a document, drafting a policy, answering a question — Copilot works well enough. What it isn’t is a command-line coding agent that can read, modify, and run code across a developer’s machine. That’s a different category of tool, and my team would get far more out of Claude Code, Codex CLI, or Gemini CLI. So the February 11 ask was narrower than it sounds: an Enterprise plan that includes a CLI coding agent, under the same kind of BAA we already have with Microsoft.
OpenAI: fast, direct, still in touch
OpenAI answered in a day. An AI Sales Representative emailed me on February 12 with ChatGPT Enterprise for Healthcare pricing: $33 per seat per month, 50-seat minimum, annual contract, Net-30. I asked whether they could flex on the minimum for a small specialty. He checked internally, came back with a polite no, then followed up again on March 20 with an approved exception — 25 seats, same per-seat price, still HIPAA-eligible with a BAA. Since then, he’s checked in.
I know exactly what OpenAI costs, what I get, and how to sign if we decide to move. That is how enterprise sales is supposed to work.
Google: meetings, partners, and “fully funded”
Google took a different path. The Gemini team wanted a meeting. Then more meetings. Then a meeting with a Google Cloud partner. Then a meeting with our dedicated account manager, a dedicated customer engineer, and a second partner director. Every conversation added people; none of them added a price.
What I eventually got, on April 14, was a fifteen-page Statement of Work from the partner for a five-week Gemini Enterprise pilot. Up to 50 test users, Microsoft 365 connectors, Entra ID federation, even a custom link to our on-prem SQL Server. The price: tens of thousands of dollars in professional services, fully offset by Google funding at project completion. Net cost to us: zero.
That $0 figure is the partner’s professional services line. Google’s Enterprise model has more separate moving parts than OpenAI’s: the Gemini seat licenses — around $35 per seat per month — are a standalone subscription, and Google Cloud consumption inside our tenant is another line item. The formal seat-license quote with any negotiated discounts is still pending from our account manager.
After four meetings and a funded pilot offer, I still don’t have what I asked for on February 11: a clean per-seat price for a small organization that wants a modest number of Enterprise seats and a BAA. I eventually had to break off the meeting cadence altogether to focus on a more pressing internal project. The pilot SOW is still on the table; the straight seat price isn’t.
Anthropic: silence
Anthropic is the one I didn’t expect to fail. Their models are the ones our team already uses through personal accounts for projects that don’t have a data component. I’d prefer to standardize on Claude. Claude Code is the CLI tool my team would be most productive with. On paper, Anthropic should have been an easy yes.
The only reply I’ve ever received from Anthropic came on February 23 — an automated response to a web-form submission, listing the non-healthcare Enterprise price: $20 per seat per month base, 20-seat minimum, with usage billed separately at API rates on top of the seat fee. Of the three vendors I evaluated, Anthropic is the only one that doesn’t bundle usage into the Enterprise seat price — so the real cost scales with volume rather than landing at a predictable flat line. The email pointed me at the self-serve purchase flow. I tried it. As soon as I selected a healthcare use case, the site redirected me back to sales. I replied to the automated thread asking for a healthcare-specific quote. Nothing. I wrote to sales@anthropic.com directly. Nothing. On April 8, I sent a LinkedIn message to a GTM lead I found on Anthropic’s healthcare page. Nothing.
To date, I have not been able to reach a single live person at Anthropic.
Why the silence?
I can only speculate. A few things are probably happening at once.
Anthropic is capacity-constrained in ways that are now public. The Register reported on April 16 that the company just restructured its Enterprise plan to eject bundled tokens and move to pure usage-based billing — the kind of change you make when demand is outrunning infrastructure. Sales reorgs follow pricing reorgs.
A 20-seat specialty practice is not the account a sales team chases when it’s triaging Fortune 500 deals. Our minimum annual commitment would be in the low five figures. A single large health system is worth seven or eight.
And “Claude for Healthcare” was announced in January. Four months later, the people I’d expect to staff that motion either don’t exist or aren’t reachable through any door I’ve knocked on.
I don’t think my experience is unusual. Anthropic’s support forum, Reddit, and the tech press have been carrying steady complaints about customer responsiveness for a couple of months now — Fortune recently described user frustration reaching the point of open revolt. What I ran into is consistent with that pattern: not a brush-off, just a company whose demand has outrun its ability to handle the humans on the other end of the contract.
What I’ll tell the committee
OpenAI has something we can act on immediately — a price, a contract, and an enrollment form. Google would almost certainly land final pricing in another meeting or two, once the team feels they’ve fully walked us through the value proposition. And I assume we’d reach someone at Anthropic eventually. The path we choose may depend less on the vendors themselves than on our own timeline and sense of urgency.