TestDriver: Safety & Security
TestDriver is an AI computer-use agent that runs end-to-end tests inside isolated cloud sandboxes.
Security Measures
- Authentication: identity via a third-party provider (OAuth / SSO); no passwords stored. GitHub access uses the GitHub App model with short-lived, per-installation tokens scoped to granted repos.
- Access control: per-team API keys, revocable at any time. Secrets and test credentials are passed as runtime env vars, never logged or committed. Inbound webhooks are verified with signing secrets.
- Isolation: every run executes in an ephemeral, single-tenant sandbox that is destroyed afterward, so no state carries between customers.
- Incident response: errors and telemetry are monitored with alerting; incidents are triaged by engineering, affected customers notified, and credentials rotated.
Data Handling
- Encryption: all data encrypted in transit (TLS/HTTPS) and at rest on managed cloud infrastructure.
- Collection: only what is needed to run tests, authenticate, and bill. Test execution data (screenshots, UI context, prompts, keystrokes, any credentials you supply) is sent to third-party AI/LLM providers to drive the agent. Account email/profile, billing data, and usage analytics are also collected.
- Privacy: data is used solely to operate the service. It is not sold and not used to train foundation models.
- Retention: sandbox compute is ephemeral; screenshots and artifacts are retained for replays/debugging and deletable on request; account and billing records are kept for the life of the account.
- Cross-border: subprocessors operate primarily in the US; data may be processed there and in other regions.
Compliance
- No SOC 2, ISO 27001, HIPAA, or FedRAMP certifications at this time.
- Payments are handled by a PCI-DSS Level 1 certified provider; TestDriver does not store raw card data.
- GDPR/CCPA data subject requests are honored. Contact support@testdriver.ai.
High-Risk Systems (EU AI Act)
TestDriver is not a high-risk AI system under the EU AI Act. It is a software-testing tool operated under human direction.
- Human oversight: tests are authored, reviewed, and triggered by developers; results are reviewed by a human. The agent acts only within the sandbox against the app under test.
- Scope: it makes no consequential decisions about individuals and serves none of the high-risk purposes in Annex III.
- Known limitations: as an AI agent, it can make mistakes (misidentify elements, take incorrect actions, or produce false pass/fail results). Output should be reviewed and is not a guarantee of correctness.