Operator.io

Docs state each instance runs in its own container with dedicated storage and AES-256-GCM encryption at rest. Data is never shared between instances or accounts. Terms say they do not access configuration data except to provide the service or comply with law. However, there is no mention of a data training opt-out policy, no employee access audit trail details, no log sanitization, and no privacy policy page is publicly accessible.

No information found about prompt injection defenses, sandboxing details for code execution, human-in-the-loop controls, memory integrity protection, or container escape prevention. The docs mention 'skills' including code execution and web search, but provide no details on how these are sandboxed or gated.

Operator manages AI inference centrally (GPT-5.2) with usage-based quotas, meaning users do not need to provide their own API keys for AI models — a genuine security advantage. However, users must provide bot tokens for Discord and Telegram. The docs do not describe how these channel credentials are stored beyond the general AES-256-GCM encryption claim.

Docs mention 'skills' that can be enabled and 'usage-based quotas and rate limiting' for AI inference. Instances stop responding when usage limits are hit. However, there is no mention of a kill switch, no behavioral monitoring, no high-risk tool gating, and no least-privilege documentation.

Terms state that after cancellation, instances are suspended but data is preserved for 14 days. No mention of backups, data export capability, disaster recovery, SLA, or uptime guarantees. No status page exists. YC-backed (W20, team of 6) provides some stability signal.

Pricing is clearly documented: Basic $15/mo, Pro $50/mo, Max $150/mo. Each plan has an AI usage quota — instances stop responding when hit, effectively a hard cap preventing runaway costs. Upgrades prorated, downgrades at end of cycle. No spending alerts or price change notification policy mentioned.

Terms state the service is provided 'as is' with no warranty. Terms can be updated at any time with no defined notification period. No mention of audit logging, incident response process, breach notification timeline, or GDPR compliance. Legal entity is Operator Research, Inc. (New York).

No information found about dependency scanning, MCP server vetting, build pipeline security, SBOM, or any supply chain security practices.

Sign-in page shows Google SSO and email/password authentication. No mention of MFA support, session management details, or security hardening. Uses Stripe for billing. Site served over HTTPS. No mention of penetration testing or independent security testing.

No information found about hallucination mitigation, approval workflows, output verification, undo/rollback capability, or transparency about AI uncertainty.