This is the policy — what we promise and what we ask of you. For how it is actually implemented (encryption, isolation, agent controls), see the Security Whitepaper. For what is certified and what is not, see Compliance Status — that page is the source of truth and we never round up.
Report a vulnerability
Email [email protected]. If you do not get a reply, email [email protected] and say it is a security report. Please include what you need to and nothing you do not:- What the issue is, and the impact you believe it has
- The minimum steps to reproduce it
- Any account, request or project identifiers involved
- How you would like to be credited, if you would like to be
We do not currently run a paid bug bounty. We will say so plainly rather than leave you guessing.
Safe harbour
If you are researching in good faith under the rules below, we will not pursue legal action and will not ask your hosting provider to act against you:- Only test against accounts and projects you own
- Do not access, modify, or retain data belonging to anyone else — if you encounter someone else’s data, stop and tell us
- No denial of service, no spam, no social engineering of our staff or users, no physical attacks
- Give us reasonable time to fix an issue before disclosing it publicly
What we are responsible for
- The platform. Our APIs, our control plane, the desktop application we ship, and the infrastructure we run for you — databases, hosting, preview links, domains.
- Isolation between customers. One organization’s data, projects and running work must never be reachable by another.
- Secrets we hold. Encrypted at rest, never returned in reversible form, and masked out of anything an AI model can see. See the whitepaper.
- Telling you when something goes wrong. See Incident notification below.
- Naming our subprocessors. The current list is published and kept current.
What stays your responsibility
Workser Code runs coding work on your computer using either a connected provider account or the built-in Workser agent. Model requests, shared project records, source sync and deployments can send content to the relevant cloud services. Review Security & approvals for the data paths and permissions.- Your machine. Disk encryption, screen lock, OS updates, and who else can use the computer the agents run on.
- Your connected model provider account. Its credentials, storage and billing follow that provider’s CLI configuration. Rotate credentials if you believe they are exposed. Workser’s built-in agent uses a separate managed-model path.
- What you approve. Consequential actions — deploying, deleting, spending, touching production data — are gated behind an approval prompt. An action you approve is an action you authorized. See Security & approvals.
- What you paste in. Anything you put into a prompt may be sent to the model provider you chose, under that provider’s terms.
- Who you invite. Organization roles (
owner / admin / member / viewer) decide who can manage billing, members, and destructive actions. Review them.
AI-specific commitments
- We do not train on your content. Not your prompts, inputs, outputs, or uploaded files — not for our general product, not for another customer — without your prior written consent.
- Review permissions are explicit. Supported read-only role configurations restrict review runs. Inspect the role’s capabilities and backing agent permissions before relying on a review boundary.
- Secrets are masked before the model sees them. Including when you ask for them explicitly.
- Protected release decisions stay with you. Desktop production deployment, live payments and protected database or bucket deletion require explicit approval. Other operations depend on role permissions and the selected autonomy policy.