Claude is good, but there is a gap it can’t close.

We use Claude, as do many of our users, and most of this page is not an argument against it.

• The document still leaves your machine. It travels to Anthropic’s infrastructure (AWS and GCP regions) for processing, and the response comes back. “Not used for training” doesn’t mean “never read by the processor”: abuse review, safety investigations, and support escalations all involve people with access.
• A contract binds the parties, not the courts. Anthropic can promise not to retain your conversations, and a US federal judge can override that promise with a preservation order. In May 2025 OpenAI was ordered to retain “deleted” ChatGPT messages over user objection. Anthropic isn’t involved in that case, but the legal mechanism behind it isn’t vendor-specific.
• The vendor can revoke access. In September 2025 Anthropic updated its terms to restrict services from entities more than 50% owned by companies headquartered in unsupported regions, and Hong Kong, Macau, and mainland China are not on Anthropic’s supported-countries list. By April 2026 Goldman Sachs had cut its Hong Kong bankers off from Claude entirely after a stricter reading of its Anthropic contract. A tool you depend on can stop working overnight for reasons unrelated to your work.

Architectural privacy, where the document never leaves the device, is the only kind a court, a regulator, or a sudden vendor policy change can’t override.

• Where the document goes. Claude: to Anthropic’s infrastructure. Muet: nowhere; it stays on your Mac.
• Training default. Claude: not used for training on paid tiers, by contract. Muet: there is no training pipeline because there is no vendor-side data.
• Retention. Claude: 30 days by default for most paid tiers, less under ZDR. Muet: none, because we have no servers that touch your work.
• Discoverable in litigation. Claude: yes, in principle. Anthropic can be subpoenaed. Muet: there is nothing to discover.
• Sub-processors. Claude: AWS, GCP, and others, listed in Anthropic’s DPA. Muet: none involved in the work itself.
• DPA required for GDPR work. Claude: yes, Anthropic’s standard DPA. Muet: not required, because Muet isn’t a processor.
• Works offline. Claude: no. Muet: yes, once activated and the model is downloaded, for up to 14 days between licence checks.
• HIPAA BAA. Claude: on Enterprise (and on Claude via Amazon Bedrock under AWS’s BAA). Muet: the BAA question doesn’t arise, because we never receive PHI.
• Price. Claude Pro: $20/mo. Claude Team: $20–$25/seat/mo (Standard), $100–$125 (Premium). Muet: $390/year per device, one annual invoice.

Its model range, long context window, integrations, and privacy posture make Claude a sensible default for the eighty per cent of professional work that is fine to send to a vendor.

Most of our paying users run both: Claude for broad work, and Muet on the laptop for the documents that have to stay put.

• Claude Free. Conversations may be reviewed for safety, so don’t use it for anything confidential. Muet covers the privileged work.
• Claude Pro ($20/mo). No training by default and 30-day retention, which is fine for most personal professional work. Muet still covers the documents you don’t want anywhere else.
• Claude Team ($20–$25/seat/mo Standard, $100–$125 Premium). Centralised admin and more usage, with the same privacy posture as Pro; Premium adds capacity, not a different privacy regime. Muet still fits confidential matters that don’t belong in the shared workspace.
• Claude Enterprise. SSO, audit logs, available BAAs, and custom data residency: the strongest enterprise terms in the category. Muet still fits on partner and associate laptops for work outside the firm’s shared deployment.