Plain thinking about on-site AI: the challenge each piece solves, who it is for, and the honest numbers underneath.
What ABA Formal Opinion 512 says about generative AI and client confidences, why boilerplate consent is not enough, and the architecture that removes the question.
Tax preparers are covered financial institutions under the FTC Safeguards Rule and must keep a written information security plan. Where a local, on-site AI fits inside that plan.
On June 10, 2025 ChatGPT and the OpenAI API were degraded for more than 10 hours. If AI is in your operational path, a provider outage is your outage. What to do about it.
On-premise AI for an 8 to 25 person office: $3,500 assessment, $18,000 to $32,000 for the first workflow installed, $6,000 to $14,000 hardware you own, $900 to $2,200 a month ongoing.
The challenge: useful AI needs your most sensitive data, and the data cannot leave. The solution: move the model into the building.
Per-token pricing punishes success. Owned hardware runs the routine tier at electricity, and four numbers decide whether it pays back for you.
Once AI is load-bearing, a provider outage is your outage. Hospitals, government, and financial firms cannot accept that, and do not have to.
What practices are actually allowed to do with AI and patient records, and the architecture that removes the hardest question.