Skip to content
IsoKron

Security

Security and disclosure

How to report a problem in isokron.ai or IsoKron’s tools, how findings in other projects are disclosed, and how client data is handled.

Report a problem

Email hello@isokron.ai about a security problem in this website (isokron.ai) or in one of IsoKron’s open-source tools. Say what you found, how to reproduce it and what it affects. Please report it privately by email rather than in a public issue.

The open-source tools this covers:

An ordinary bug in a tool, with no security impact, can go straight to that tool’s issue tracker on GitHub. Testing of this website stays within the site terms: no attempts to disrupt or overload it. The same contact is published, machine-readable, at /.well-known/security.txt.

Disclosure policy

When IsoKron finds a problem in someone else’s MCP server or software:

  1. Correctness bugs in public servers go to public issues on the project’s own tracker.
  2. Anything with security impact goes through the project’s private reporting channel, with a 90-day default embargo.
  3. Client findings are never disclosed without the client’s written consent.

Bugs we find in third-party dependencies of a client’s server go to their maintainers only after we coordinate with the client.

Every public filing is listed, with its status, on the track record. An open issue is a report, not a confirmed vulnerability, until the maintainers say so.

Client access and data

  • Test access only. We work on test tenants with tokens scoped to them, never on production data or production tenants, and we never ask for production credentials or customer data.
  • No secrets by email. Never send tokens, passwords or keys by email or through the request form; we agree with you how to share access first.
  • Tokens end with the engagement. At the end of an engagement we revoke or delete the tokens we were given, and we ask you to revoke them on your side too.
  • Evidence is deleted. Reproduction scripts, logs, tool outputs and reports are kept for the period set in the engagement agreement, then deleted.
  • AI model providers. Test data from your test tenant, tool outputs and session transcripts may be processed by the AI model providers we use, under their API terms. This is one more reason testing never uses production data.
  • Private by default. Reports are private, and we sign a reasonable NDA on request. Nothing about an engagement is published without your written consent.

The full policy is on the privacy page; each engagement’s own written terms set the details, as the terms explain.