macOS Detailed Score Privacy Policy with AI Details (EN)

By reporting a detailed score, you agree to share the following information with EDAMAME:

• Your machine unique identifier
• Your operating system name and version
• Your public IPv4 address and/or IPv6 address
• Your MAC address if available
• Your peer IDs for your VPN or ZTNA connections if available
• The domain you are connected to
• Your username in that domain
• Your score as a single numerical value
• Your score as a detailed vector of boolean values resulting on the following security checks:
  • EDAMAME helper inactive
  • Response to ping enabled
  • MDM profiles installed
  • JAMF remote administration enabled
  • Wake On LAN enabled
  • Manual Appstore updates
  • Local firewall disabled
  • Automatic login enabled
  • Remote login enabled
  • Remote desktop enabled
  • File sharing enabled
  • Remote events enabled
  • Corporate disk recovery key
  • Disk encryption disabled
  • Unsigned applications allowed
  • Manual system updates
  • Screen lock disabled
  • No antivirus enabled
  • No password manager installed
  • System Integrity Protection disabled
  • Guest account enabled
  • Root user enabled
  • Unprotected system changes
  • Potentially compromised email address
  • Unverified or unsafe network environment
  • Unverified or unsafe services exposed to the LAN
  • Unverified or anomalous traffic
  • Unreviewed vulnerability findings
  • Behavioral divergence detected
  • Escalated actions pending review
  • Your OS is not up to date
  • Chrome browser not up to date
  • Business rule not respected
  • CLI not restricted for standard users
  • Cursor agent unsecured
  • Claude Code agent unsecured
  • Claude Desktop agent unsecured
  • OpenClaw agent unsecured
  • AI agent with high host blast radius
  • AI agents run without a governance harness
  • Agent escapes its governance harness boundary
  • AI agent exposes an unprotected MCP server
  • Codex CLI agent unsecured
  • Hermes agent unsecured
• The details of each failing AI agent security check listed above:
  • For every AI coding agent EDAMAME supports, whether it is installed on this machine and whether its transcript observer is running
  • The name of the agent a failure belongs to, for example cursor or claude_code
  • The name of the governance harness that agent declares, for example nono or srt
  • The name of the risk amplifier that fired, for example passwordless_root, critical_subprocess or secret_exposure
  • The file name, without its path or its arguments, of a sensitive program the agent launched, for example ssh
  • The configured name of an MCP server found to be exposed, for example gojiberry, together with the exposure rule that fired, for example mcp_public_no_strong_auth. MCP servers that are not exposed are never named
  • The category of a secret found in the agent transcript, for example aws_credentials, never the secret itself

Agent transcripts, prompts, model responses, file contents, command arguments, environment variable values and secret values are never reported.

This information is used solely by EDAMAME and is not shared with any third party.

This information is gathered using a public "threat model" that is guaranteed not to violate your privacy.

The threat model can be seen at https://github.com/edamametechnologies/threatmodels/blob/main/threatmodel-macOS.json.

The threat model wiki can be seen at https://github.com/edamametechnologies/threatmodels/wiki/threatmodel-macOS-EN.

If you do not agree with this policy, please do not report your score.