Anthropic Claude Privacy Guide

Privacy score: 82/100. Transparency score: 88/100. Overall 72/100 (Teen Approved), reviewed 2026-07-15.

What the vendor documentation says

Anthropic publishes a detailed Privacy Policy and a Trust Center covering SOC 2 Type II, ISO 27001, ISO 27701, ISO 42001 (AI management), and HIPAA support on eligible plans. By default Anthropic does not train its models on consumer chat inputs or outputs unless the user opts in; API/commercial data is not used for training by default and is subject to a Zero Data Retention option for enterprise. Consumer ToS requires users to be 18+, so COPPA does not apply because the service is not directed to children under 13 — but that same fact makes the product out-of-scope for direct K-12 student use. GDPR: Anthropic offers a DPA and lists an EU representative; data may be processed in the US with SCCs. FERPA: no student-record processing agreement is offered for K-12; Claude for Education targets higher-ed institutions. EU AI Act: Anthropic publishes model cards and a Usage Policy consistent with GPAI transparency obligations, and Claude models are covered by Anthropic's Responsible Scaling Policy.

Controls and safeguards on the data path

Safeguards target adult users: no parental controls, no per-child accounts, no time limits, no built-in profanity filter, no simplified reading level, and no companion-style relationship guardrails specific to minors. The Usage Policy prohibits use by minors on consumer Claude.ai and prohibits high-risk use cases (medical, legal, election, etc.) without a human in the loop. On the API, developers are responsible for age gating, safety classifiers, and downstream use.

Open privacy concerns

• Consumer ToS requires users to be 18+ — not appropriate for direct K-12 student use.
• No COPPA-compliant K-12 offering and no FERPA school-official DPA marketed for K-12 districts.
• No child-facing safeguards: no parental controls, no kid-mode, no session limits, no simplified reading level.
• Adult-oriented content (mature themes, ambiguous mental-health discussions) is gated for adults, not blocked for minors who may access it anyway.
• Powerful capabilities (code, essays, research) raise academic-integrity risk when used without teacher framing.

Questions to ask before you sign

• Is there a signed agreement that names the district as the data controller and Anthropic as processor for student records?
• Is there any K-12 age-gated deployment mode, or only the 18+ consumer and higher-ed products?
• Can we disable training and enable zero data retention on the plan we would purchase?
• What staff-use guardrails do you recommend for districts that permit teacher-only access?

District posture we recommend

Policy recommendation for K-12 administrators: block direct student access to Claude.ai on managed networks and devices; permit staff use for administrative and instructional-prep purposes only, under an AUP that requires (1) no student PII in prompts, (2) human review of all outputs, and (3) disclosure when AI was used to generate materials. Do not sign an enterprise agreement expecting FERPA coverage — Anthropic's current DPA and Trust Center commitments are enterprise-grade but not marketed as FERPA school-official terms for K-12. Reassess if Anthropic launches a K-12 product with a signed DPA and age-appropriate defaults.

What this article does not claim

Findings labelled as verified come from SafeGradeAI's published evaluation of this tool and the vendor's own public documentation as of the review date above. Anything framed as a recommendation is our professional guidance, not a statement of fact about the product. Vendor policies, model versions, and default settings change frequently — re-check the vendor's current documentation before making a district-level decision, and treat any AI output as a draft to verify.

---

This guide is built on the SafeGradeAI evaluation of Anthropic Claude. Read the full Anthropic Claude evaluation and rubric scores.