Security review
Use this page as the starting point for a security review of AgentPaaS. It links to the current control descriptions, evidence paths, and stated limits without replacing your review of agent code, dependencies, policies, destinations, or provider terms.
Security posture
- Threat model explains the boundaries, enforcement tiers, and claims AgentPaaS does not make.
- How enforcement works shows how signed bundles, default-deny egress, gateway enforcement, brokered credentials, and audit records fit together.
- Architecture describes what runs locally and in Cloud, request flow, authority boundaries, and responsibility boundaries.
Credentials and data
- Credentials and secrets describes storage, gateway brokering, policy bindings, and credential audit records.
- Data handling and LLM providers explains where prompts and tool data can go, US Cloud residency, and which provider settings remain your responsibility.
Governance and audit
- Governance and audit explains how AgentPaaS connects workload identity, policy decisions, run lineage, and audit evidence.
- Runs and Audit Logs covers run status, logs, events, and operational review.
- Audit export and verification covers signed, hash-chained JSONL exports and what verification can detect. Tail deletion remains a documented limitation.
Evidence and attestations
- Compliance and attestations states the current SOC 2 status, sole infrastructure subprocessor, US data residency, and penetration-testing status.
Read the limits
- Known limitations records unsupported workflow paths, HTTP/HTTPS scope, default-tier assurance, local-mode trust, signing gaps, and audit limits.
The Cloud default tier does not claim substrate-enforced isolation. The paid, on-request high-assurance tier provides that additional substrate boundary. AgentPaaS does not claim prompt-injection prevention or SOC 2 certification.