Security
A security product for agents will be attacked by agents. Here is how to reach us, and what we hold ourselves to.
Report a vulnerability
Email security@securemyagent.ai. Include steps to reproduce and the affected URL or endpoint. We acknowledge reports and keep you updated until the fix ships.
- Good-faith research is welcome.
- Do not read, change or delete another tenant's data. Use your own agent.
- Do not run denial-of-service tests against production.
- A bug bounty opens with the public launch.
Machine-readable: /.well-known/security.txt
Report a leaked key or a malicious skill
Leaked an sma_ key? Rotate it through the API or email us. Found a malicious skill or MCP server? Use POST /v1/reports or the same address.
How SMA protects itself
- No secrets in scan uploads
- Uploads carry fingerprints and redacted locations. A server-side scanner rejects any key pattern.
- Keys are hashed
- An sma_ key is shown once. SMA stores a keyed hash with a pepper in AWS KMS and never logs the key.
- Crawled code is data
- It is never executed. It is fetched through an SSRF guard and unpacked in a Lambda with no network.
- Signed releases
- The CLI, skill and MCP server are signed with Sigstore. SLSA provenance and checksums are published.
- Tenant isolation
- Every record is partitioned by agent or owner. IAM conditions enforce it.
- SMA scores itself
- SMA runs its own Checkup, and ClawGRC maps SMA to SOC 2 and ISO 27001.
Full list: how SMA protects itself.