Claude security settings
Claude Security Settings: Top 10 for Admins
Claude comes as a personal app, a company workspace (Team or Enterprise), the Claude Console for building apps and agents on your computer (Claude Code and Claude Desktop). Each has its own settings. These are the 10 settings that matter most, where to find each one and what good looks like.
In short: start with (1) SSO set up and required, (2) Domain verified; new personal orgs blocked and (3) Connectors allowed and their tool permissions, then work down the list below.
By Jon Rose, Founder & Managing Partner, IOmergent. Vendor facts as of October 2026. Not sure which version you have? See which AI account is which.
The top 10 Claude security settings
Most important first. Our free AI Hardening Check walks you through these and the other 55, with direct links, inside your own AI assistant.
-
1SSO set up and required
- Where to find it
- Organization settings > Organization and access > Authentication: 'Require SSO for Claude' (and 'for Console'); provisioning Invite-only or JIT
- Good looks like
- SSO configured and required for Claude (Console sign-in is ANT-18)
- Why it matters
- One identity to disable when someone leaves
-
2Domain verified; new personal orgs blocked
- Where to find it
- Organization settings > Organization and access > Domains (verify via DNS TXT) and Security > 'Restrict organization creation'
- Good looks like
- Company domain verified and organization creation restricted
- Why it matters
- Stops staff creating personal Claude / Console orgs with work email
-
3Connectors allowed and their tool permissions
- Where to find it
- Organization settings > Connectors (allowed list, member requests); each connector > Tool permissions (Always allow / Needs approval / Blocked)
- Good looks like
- Only connectors in use and approved; write tools 'Needs approval' or 'Blocked'; unused ones removed (owners enable them for the org)
- Why it matters
- Connectors let Claude read and act in other systems
-
4Code execution, file creation and Skills
- Where to find it
- Organization settings > Skills (newer UI: Plugins & skills > Policy; older UI: Capabilities)
- Good looks like
- A deliberate on/off decision, recorded
- Why it matters
- On by default on Team; runs code in Anthropic's sandbox
-
5Chat, session and project sharing
- Where to find it
- Organization settings > Data and privacy > Sharing: Share chats externally, Share chats that use connectors, Share cloud sessions, Share projects / Public projects. The page shows 'Default' or 'Set by admin', not on/off: read each toggle
- Good looks like
- Share chats externally off; connector chats and cloud sessions shared only if needed; project sharing a deliberate choice
- Why it matters
- External chat sharing gives outsiders a copy; connector chats can quote email, Drive or repo content
-
6Data retention and export
- Where to find it
- Organization settings > Data and Privacy (custom retention is Enterprise only; export is primary owner only)
- Good looks like
- Enterprise: retention period set; Team: note that chats are kept until deleted
- Why it matters
- Defines how long chats with company data exist
-
7Managed settings deployed
- Where to find it
- managed-settings.json in /Library/Application Support/ClaudeCode/ or plist com.anthropic.claudecode (macOS), /etc/claude-code/ (Linux), C:\Program Files\ClaudeCode\ or HKLM\SOFTWARE\Policies\ClaudeCode (Windows; HKCU does not enforce); or claude.ai Organization settings > Claude Code > Managed settings
- Good looks like
- Deployed by MDM on every machine. Server-managed settings alone skip local Cowork, Console-key, Bedrock, Vertex and custom base URL sessions
- Why it matters
- Managed settings outrank every user and project setting
-
8Bypass permissions mode disabled
- Where to find it
- Managed settings: permissions.disableBypassPermissionsMode
- Good looks like
- "disable"; defaultMode is never bypassPermissions
- Why it matters
- Bypass mode skips every permission prompt
-
9MCP server allowlist
- Where to find it
- Managed settings: allowedMcpServers, deniedMcpServers, allowManagedMcpServersOnly; or managed-mcp.json, which can only be deployed as a file, not through server-managed settings
- Good looks like
- Allowlist set and locked
- Why it matters
- If unset, any MCP server can be added
-
10API keys: owner, expiry, unused keys
- Where to find it
- Settings > API keys (Disable is reversible, Delete is permanent)
- Good looks like
- Every key has an owner and an expiry; unused keys disabled
- Why it matters
- Leaked or orphaned keys are the main API risk
Claude security settings: common questions
- What are the most important Claude security settings?
- Start with (1) SSO set up and required, (2) Domain verified; new personal orgs blocked and (3) Connectors allowed and their tool permissions. All 10 are listed on this page, most important first, with where to find each one and what good looks like.
- Does Claude Team train on our data?
- No. Anthropic doesn't train on Claude Team or Enterprise data under its commercial terms. On personal plans it's each person's choice, under 'Help improve Claude'.
- How long does Claude Team keep chats?
- Deleted chats are purged within 30 days. Enterprise can set custom retention, with a 30 day minimum.
- Does Claude Team support SSO and SCIM?
- Team supports SSO but not SCIM. With a verified domain, 'Restrict organization creation' stops new personal orgs on your domain. Enterprise adds SCIM and domain capture.
- How can I check all the Claude settings quickly?
- Add the free IOmergent AI Hardening Check to your AI assistant and ask: "Run the IOmergent AI Hardening Check for our Claude Team workspace and Claude Code." It walks you through these and the other 55, with direct links, and nothing leaves your machine without your OK.
Want an expert review?
A self-check can't see which AI apps hold OAuth access to your email, files and code, or confirm the settings match what was reported. Our AI Security Assessment does both, with read-only connections and a live review.
See the AI Security Assessment