Shadow AI · Tool risk profile

Claude.

by Anthropic · generative ai · Verified April 19, 2026

Vendor site

Base risk

3.2/ 5

LowMediumHighCritical

Anthropic does not train on customer inputs by default across consumer or enterprise tiers, which puts Claude in a stronger default privacy posture than most general-purpose chatbots. Risk concentrates in the new Computer Use and agent features, which can take real actions in browsers and applications, and in the long-context window which makes large-scale document leakage easier in a single chat. Enterprise tier adds SSO, audit logs, and a no-retention option suitable for regulated workflows.

Tier comparison

Same logo. Very different risks.

  • Free

    medium
    Trains on inputs
    No
    Retention
    30 days
    SSO
    No
    Admin controls
    No

  • Paid · consumer

    medium
    Trains on inputs
    No
    Retention
    30 days
    SSO
    No
    Admin controls
    No

  • Enterprise · team

    low
    Trains on inputs
    No
    Retention
    30 days
    SSO
    Yes
    Admin controls
    Yes

FAQ

Questions teams ask about Claude.

Does Claude train on customer chats?

No — Anthropic does not use prompts or completions to train its base models by default on any tier. Training data uses are documented in their Usage Policy.

Can Claude be used with PHI or financial customer data?

Anthropic offers BAAs for Claude Enterprise and the Claude API; consumer tiers are not appropriate for PHI without one.

What is the residency story for Claude?

Anthropic processes data primarily in US regions; EU data residency is available on the Claude Enterprise plan via Amazon Bedrock or Google Vertex AI.

Audit your shadow AI

Is Claude live in your org
alongside tools IT doesn’t know about?

Run a free 12-minute audit to surface every shadow AI tool on your network, score the risk, and walk away with a block-list your IT team can import.

Start the free auditMethodology

Buzzi.ai publishes tool risk profiles for informational purposes only. Always validate terms with the vendor before operational decisions.