Anthropic Max plan ToS — productized/agentic use vs personal use, and the real enforcement record
The question
What does Anthropic's Max plan Terms of Service actually say about productized / agentic use vs personal use, and are there documented cases (forum posts, support-thread reports, founder anecdotes) of Max (or Claude subscription) accounts being throttled, rate-limited, or terminated for crossing the personal-vs-commercial line? Context: RDCO runs an always-on single-founder "Ray" COO agent on a Claude subscription substrate; being wrong here costs an account termination plus reputation, so it is high-stakes and load-bearing for Ray's cost economics.
What we already know (from the vault)
- [[2026-05-05-personal-license-boundary]] — founder's position memo: "Anthropic Max Plan is a personal-use license... The Max Plan covers Ray-as-personal-COO, not Ray-as-multi-tenant-SaaS." Flip to API / per-seat commercial only if Ray serves a paying customer, a multi-user instance, external distribution, or a teammate/co-founder. Decision rule: "single-founder personal use → proceed; crosses multi-user / shared-product / external-distribution threshold → STOP and surface."
- [[2026-04-27-indy-dev-dan-maximize-claude-code-subscription]] — the operating test: "one human, one subscription, one beneficiary" and "am I the only human whose work these agents are running?" Bins uses into safe / controversial / bannable. Our cron + dynamic-loop pattern maps cleanly to the SAFE tier because the founder is the sole beneficiary. Names the actual bannable vectors: shipping a product on Pro/Max OAuth, multi-tenant SaaS logging in on users' behalf, token extraction from
~/.claude.json/keychain, wiring OAuth into third-party harnesses. - The same memo already flags the drift risks: a Claude-backed endpoint on an RDCO surface, an auto-reply-to-subscribers pipeline, or any non-founder request routed through the Max subscription all cross into product territory and require an API key.
- [[2026-07-02-claude-sandboxes-security-pricing-vs-rdco-substrate]] and [[2026-06-03-claude-api-sdk-capability-map-rdco-use-cases]] hold the adjacent substrate-cost and commercial-API-path context.
What the web says
- Anthropic's Consumer Terms (rollout Aug 28 2025; acceptance deadline Oct 8 2025) explicitly cover Free/Pro/Max "including when they use Claude Code from accounts associated with those plans," and explicitly do NOT apply to Commercial Terms services (Claude for Work/Government/Education, or API use including Amazon Bedrock and Google Vertex AI). So the Consumer→Commercial split IS the personal-vs-commercial boundary. (anthropic.com/news/updates-to-our-consumer-terms)
- The Usage Policy update (effective Sept 15 2025) added an agentic-use section plus a Help Center article on "how our Usage Policy applies to agentic use" with concrete prohibited-activity examples; the operative language lives in the AUP at anthropic.com/legal/aup, not in the news post, which itself draws no personal-vs-commercial line. (anthropic.com/news/usage-policy-update)
- Practitioner synthesis: the Claude Code CLI is Anthropic's own first-party product and is permitted "whether you run it locally or on a VPS, whether it's for personal or business use." The Consumer ToS automation prohibition has two explicit exceptions: via an API key, or where Anthropic otherwise explicitly permits it (Claude Code CLI). For anything built for a business, the zero-ambiguity path is an API key under Commercial Terms. (autonomee.ai)
- Documented enforcement (the only real cases found): in January 2026 Anthropic suspended accounts running OAuth-token-extraction tools (OpenClaw, OpenCode, Roo Code, Goose) that spoof the Claude Code harness. The line drawn: "calling
claude -p(the real CLI) = allowed. Extracting OAuth tokens to use in a third-party API client = banned." No documented case of a solo user's local or always-on stock-CLI automation being banned. (autonomee.ai; corroborates the IndyDevDan OpenClaw timeline) - Usage-limit mechanics (2026): a dual-layer system — a 5-hour rolling window (doubled May 6 2026) plus weekly caps (Max carries two: all-models and Sonnet-only), temporarily +50% through July 13 2026. Max 5x = $100/mo, Max 20x = $200/mo. (morphllm.com, truefoundry.com)
- KEY change: from June 15 2026, non-interactive subscription usage (Agent SDK,
claude -p, GitHub Actions, third-party apps authenticating with your subscription) stopped competing with interactive sessions and now draws from a SEPARATE monthly credit ($20 Pro / $100 Max 5x / $200 Max 20x). This is Anthropic building explicit product support for exactly the subscriber-owned automated-agent pattern RDCO runs. (morphllm.com / search synthesis)
Convergences and contradictions
- Strong convergence: the founder memo, IndyDevDan's rule, and the actual ToS agree that the flip line is Consumer→Commercial and that it trips on the same triggers (serving a non-founder, shipping a product, multi-user/shared-product). The memo's decision rule and "one human, one subscription, one beneficiary" are the same test in different words, and both hold under the current (2026) terms.
- Reassuring gap: no forum/support/anecdote evidence of a solo, local, always-on stock-CLI user being terminated. Every real suspension traced to OAuth-token extraction or harness-spoofing, which RDCO explicitly does not do.
- One evolution to note: Feb-2026 reporting said the Agent SDK required API-key auth and could not use Pro/Max OAuth, whereas the June-15-2026 usage change describes Agent SDK "authenticating with your subscription" drawing a separate credit. Policy loosened toward metered subscription non-interactive use between those dates. Not load-bearing for RDCO, since Ray runs stock
claude/claude -p, not the raw Agent SDK.
Synthesis for RDCO
The verdict is that the always-on single-founder Ray setup sits safely inside the Consumer/subscription terms as currently written, and the June 15 2026 change strengthens rather than threatens that position. Ray runs stock Claude Code (claude, claude -p via cron/launchd), which is the exact first-party product Anthropic explicitly permits for automated and non-interactive use, and the founder is the sole beneficiary. That satisfies both the actual Consumer→Commercial boundary and the "one human, one subscription, one beneficiary" test. The existing "personal-use is fine for solo-founder RDCO" memo still holds; nothing in the 2026 ToS invalidates it, and no migration to the API is forced today.
The cost economics do not push toward a commercial migration either. The June 15 non-interactive separate credit ($100/mo on Max 5x, $200 on Max 20x) means Anthropic now meters exactly RDCO's cron/loop pattern and gives it dedicated headroom instead of forcing it to compete with the founder's interactive Claude usage. The practical watch item is a cost-planning question, not a ToS one: if the autonomous loop's non-interactive spend exceeds that monthly credit, the overflow behavior (hard throttle vs pay-as-you-go vs interactive-pool fallback) needs verification so uptime is predictable.
The concrete triggers that would flip RDCO to Commercial/API are the same ones the memo already lists: (a) any non-founder human's request routed through the Max subscription for Claude to act on — note the Discord external-request path currently only tags the founder rather than doing work for outsiders, and it must stay on that side; (b) a Claude-backed endpoint exposed on any RDCO surface (chat widget, "ask the agent," inbound-email auto-reply); (c) a multi-tenant or shared-product instance of any Ray-derived tool; (d) the first paying customer, teammate, or co-founder who uses Ray. When any of these fires, the path is an API key under Commercial Terms (see [[2026-06-03-claude-api-sdk-capability-map-rdco-use-cases]]), not stretching the subscription.
The things RDCO must never drift into are precisely what actually gets accounts suspended: wiring OAuth into third-party harnesses (OpenClaw/OpenCode/Goose/Roo), extracting tokens from ~/.claude.json or Keychain to re-inject elsewhere, or spoofing the Claude Code harness. Because RDCO uses stock CC, it is clear. One caveat worth respecting: the operative personal-vs-commercial and prohibited-agentic-activity language lives in anthropic.com/legal/aup and the agentic-use Help Center article, not in the marketing news posts read here, so a periodic verbatim re-read of those two canonical pages is the right hygiene given the stakes.
Open follow-ups
- What is the exact overflow behavior when the Max non-interactive monthly credit ($100 / $200) is exhausted — hard throttle, silent fallback to the interactive pool, or pay-as-you-go overage? Load-bearing for the always-on loop's uptime and monthly cost predictability.
- Verbatim read of anthropic.com/legal/aup plus the agentic-use Help Center article (support.anthropic.com/en/articles/12005017) to capture the actual prohibited-agentic-activity examples; the news posts do not contain the operative clauses.
- Does the Discord non-founder-request path (currently tag-founder-only) ever perform Claude work on an outsider's behalf? If so it edges toward the controversial line; confirm the skill never routes an outsider's task through the subscription.
- Is Ray's live auth path confirmed to be OAuth/subscription (
api_key_source: none,five_hourrate-limit events) rather than silently billingANTHROPIC_API_KEY? One-time verification per the IndyDevDan audit item.
Related
- [[2026-05-05-personal-license-boundary]]
- [[2026-04-27-indy-dev-dan-maximize-claude-code-subscription]]
- [[2026-07-02-claude-sandboxes-security-pricing-vs-rdco-substrate]]
- [[2026-06-03-claude-api-sdk-capability-map-rdco-use-cases]]
Sources
- Vault: ~/rdco-vault/01-projects/bookstore-for-agents/2026-05-05-personal-license-boundary.md — [[2026-05-05-personal-license-boundary]]
- Vault: ~/rdco-vault/06-reference/2026-04-27-indy-dev-dan-maximize-claude-code-subscription.md — [[2026-04-27-indy-dev-dan-maximize-claude-code-subscription]]
- Vault: ~/rdco-vault/06-reference/research/2026-07-02-claude-sandboxes-security-pricing-vs-rdco-substrate.md — [[2026-07-02-claude-sandboxes-security-pricing-vs-rdco-substrate]]
- Vault: ~/rdco-vault/08-tooling/2026-06-03-claude-api-sdk-capability-map-rdco-use-cases.md — [[2026-06-03-claude-api-sdk-capability-map-rdco-use-cases]]
- Web: Anthropic, "Updates to Consumer Terms and Privacy Policy" — https://www.anthropic.com/news/updates-to-our-consumer-terms
- Web: Anthropic, "Usage Policy Update" — https://www.anthropic.com/news/usage-policy-update
- Web: Anthropic Acceptable Use Policy (canonical, not yet read verbatim) — https://www.anthropic.com/legal/aup
- Web: autonomee.ai, "Is This Allowed? Claude Code Terms of Service Explained" — https://autonomee.ai/blog/claude-code-terms-of-service-explained/
- Web: morphllm.com, "Claude Code Usage Limits (2026)" — https://www.morphllm.com/claude-code-usage-limits
- Web: truefoundry.com, "Claude Code Rate Limits & Usage Quotas Explained (2026)" — https://www.truefoundry.com/blog/claude-code-limits-explained