Claude Trust Audit: The Privacy-Friendly AI That Now Trains on Your Chats Unless You Say No

What it actually costs

Free is $0, with chat on web, desktop and mobile, web search and file creation. Pro is $20 a month billed monthly, or $17 a month on an annual subscription — $200 up front. Max is sold as “from $100”: 5x is $100 a month and 20x is $200 a month for web subscriptions, billed monthly only, with no annual discount. Team and Enterprise are seat-based plans for organisations.

Each step buys mostly headroom. Pro gives at least 5x more usage per five-hour session than Free, plus Claude Code, Projects and more models. Max gives 5x or 20x Pro’s usage per session, higher output limits, early access and priority access. Its listed $100–$200 in monthly API credits are separate from the chat allowance, unusable for Claude Code, and expire each cycle.

The traps are in the limits:

• No fixed message count. Anthropic says so: how much you can do depends on conversation length and complexity, the model and the features you use. Sites quoting an exact figure are guessing.
• Two clocks run at once. Every plan has a limit resetting on a rolling five-hour window; paid plans add weekly limits on top. The five-hour reset does not fix a weekly block.
• One shared pool. Web, desktop, mobile and Claude Code all draw from the same allowance, so a heavy morning of coding can block your chat that afternoon. Free has no Claude Code at all.
• Going over costs extra. At a limit you can wait, upgrade, or turn on usage credits at standard API rates — the flat fee can quietly become a metered bill. Even the top Fable model is rationed: paid credits only on Pro, capped at 50% of weekly limits on Max.

Prices exclude tax, and the page states that “price and plans are subject to change at Anthropic’s discretion.”

Your data: what the fine print actually says

Before August 2025, Anthropic did not use consumer conversations to train Claude, unless a user submitted feedback. On August 28, 2025, it announced updates to its Consumer Terms and Privacy Policy covering Free, Pro and Max, including Claude Code on those accounts.

The announcement presents this as a choice: new users select a preference at sign-up, existing users got a pop-up, and the setting can be changed any time in Privacy Settings (toggling off “Help improve Claude”). The legal wording is blunter: the Consumer Terms say Anthropic may use your materials, including training its models, “unless you opt out of training through your account settings,” and the pricing page lists training for every consumer plan as “Opt-out.”

Retention follows the toggle. Allow training and retention extends to five years for new or resumed chats and coding sessions — the same five years applies to feedback you submit. Decline it and you keep the existing 30-day retention. Deleted conversations are not used for future training and leave the back-end within 30 days, per the Privacy Policy.

The opt-out has holes Anthropic states openly: even opted out, your material will still be used for model improvement when flagged for safety review, or when you reported it via feedback. Anthropic does say it does not sell users’ data.

The sharpest divide is consumer versus commercial. The 2025 changes do not apply to services under the Commercial Terms — Claude for Work, Education or Government, or API use via Bedrock and Vertex AI. Max at $200 a month buys no stronger training protection than Free.

Refunds & cancellation

The Consumer Terms state: “Except as expressly provided in these Terms or where required by law, all payments are non-refundable.” The pricing page softens it only slightly: payments are “generally non-refundable.”

• Cancelling is not a refund. Fees will not be refunded, but access continues to the end of the term paid for. On annual Pro, that can mean walking away from prepaid months.
• The exceptions are mostly legal, not generous. Customers in the European Economic Area and the UK may get a refund within 14 days of purchase, prorated by use. Elsewhere the route is an eligibility check — “Get help,” then “Claude Refund Request” — and eligibility is Anthropic’s call.
• Where you bought it decides who refunds you. App Store subscriptions can only be refunded by Apple, at Apple’s discretion; Google Play needs a support eligibility check; no refund can be processed during a pending bank dispute. Termination for a terms breach means no refund at all.

Cancelling itself is clean: any time in Settings > Billing, keep the plan to the end of the period, cancel at least 24 hours before renewal. App-store subscribers cancel through Apple or Google instead. Cancelling does not delete your data — chats, projects and files stay as the account drops to Free.

Company risk

Shutdown risk is low. In September 2025 Anthropic closed a $13 billion Series F at a $183 billion post-money valuation, led by ICONIQ, Fidelity and Lightspeed, and said run-rate revenue had grown from about $1 billion at the start of 2025 to over $5 billion by August. Amazon is a major backer too: reporting on its November 2024 announcement put its total investment at $8 billion, with AWS as primary training partner.

The real risk is pricing power and repackaging — already used once, in the 2025 training change. The terms also let Anthropic change subscription features, increase or decrease capacity limits, and revise the terms at its discretion. Fee increases need 30 days’ notice and do not apply during your current term, but apply at renewal unless you cancel.

Who should / shouldn’t buy it

Buy it if: you use Claude daily, keep hitting the free limits, and want Claude Code included — Pro at $20 a month, billed monthly, is the sensible entry point. Buy only after turning off the training toggle on day one. Max makes sense solely for people repeatedly stopped by Pro’s limits; it buys capacity, not better privacy.

Skip it if: you need confidential handling of client, legal, medical or proprietary material on a personal account — the consumer terms, opt-out or not, are the wrong vehicle. Skip annual billing if the no-refund rule bothers you, skip Max if you expect an unlimited plan (there isn’t one), and skip paying if five-year training retention is a deal-breaker you will not manage yourself.

Bottom line

Trust score 6/10. Claude’s terms are clearer than most, cancellation is clean, and the company is well-funded. But a trust audit judges the defaults: training is opt-out rather than off, retention stretches to five years, the opt-out has safety and feedback exceptions, usage is capped by numbers Anthropic declines to publish, and money is gone once paid. Pro, billed monthly, toggle off, is defensible. Anything more, or anything sensitive, deserves a pause.

How we audit: we have not purchased or hands-on tested Claude. This audit is written from Anthropic’s own pricing page, help centre and terms, plus reputable published reporting, checked in October 2026. Prices and terms change; the links below show exactly where each claim comes from.

Sources

• Claude plans and pricing page (Free/Pro/Max prices, annual vs monthly billing, 5x/20x usage, five-hour and weekly limits, shared usage pool, no fixed message count, tax and change-at-discretion wording, refund and cancellation FAQs): https://claude.com/pricing
• Anthropic Consumer Terms of Service (training unless you opt out, safety and feedback exceptions, all payments non-refundable, auto-renewal, cancellation without refund, fee-increase notice, termination without refund for breach): https://www.anthropic.com/legal/consumer-terms
• Anthropic Privacy Policy (training use of inputs and outputs unless you opt out, flagged-for-safety and reported-material exceptions, deleted conversations removed from back-end within 30 days, no sale of personal data): https://www.anthropic.com/legal/privacy
• Anthropic announcement, Updates to Consumer Terms and Privacy Policy, August 28, 2025 (applies to Free/Pro/Max including Claude Code, not to commercial/API services, five-year retention if training is allowed, 30 days if not, five years for feedback, changeable in Privacy Settings): https://www.anthropic.com/news/updates-to-our-consumer-terms
• Claude Help Center, Request a refund for a paid Claude plan (EEA/UK refund within 14 days prorated by use, Apple-only refunds for App Store purchases, Google Play eligibility checks, no refunds during a pending bank dispute): https://support.claude.com/en/articles/12386328-request-a-refund-for-a-paid-claude-plan
• MacRumors reporting on the August 2025 change (opt out at sign-up or via Settings > Privacy, previous position of not training on consumer conversations except feedback): https://www.macrumors.com/2025/08/28/anthropic-claude-chat-training/
• IntuitionLabs, Claude Max plan pricing (Max 5x $100/month and Max 20x $200/month for web subscriptions, monthly billing, API credits expire at the end of each billing cycle and do not cover Claude Code): https://intuitionlabs.ai/articles/claude-max-plan-pricing-usage-limits
• Anthropic announcement, Series F financing ($13 billion raised at $183 billion post-money valuation, run-rate revenue from about $1 billion to over $5 billion in 2025, over 300,000 business customers): https://www.anthropic.com/news/anthropic-raises-series-f-at-usd183b-post-money-valuation
• PYMNTS reporting on Amazon’s investment (additional $4 billion taking Amazon’s total to $8 billion, AWS as primary training partner, Amazon remaining a minority investor): http://pymnts.com/news/investment-tracker/2024/amazon-increases-total-investment-artificial-intelligence-startup-anthropic-8-billion-dollars/?ref=biztoc.com