Home / Threat model
ObliviOS Threat Model
Who could attack ObliviOS, along which path, and what stops, detects and repairs it.
1Scope and objectives
This model covers ObliviOS-owned additions, its GrapheneOS-derived base, release operations, and supported Pixel devices. It does not claim to defeat a fully compromised hardware root of trust, unknown vulnerabilities, or network-level denial of service. Its objectives are confidentiality of user data and keys, integrity of boot/update/policy decisions, availability of essential phone functions, isolation between apps and Worlds, and accountable remediation.
2Assets
- User data and decryption keys
- Never exposed to an untrusted app, service, model, or operator.
- Authentication secrets
- Hardware-backed where supported and never exported by ObliviOS.
- App and World isolation
- No access across boundaries without an explicit Android/ObliviOS sharing path.
- Network policy
- Only typed, authorized policy operations mutate it; every mutation is auditable.
- Signed update trust
- Only ObliviOS-authorized verified artifacts install; downgrade attacks are rejected.
- Recovery authority
- Restores a trusted state without weakening verified boot or disclosing protected data.
- AI broker authority
- Limited to validated typed requests; an untrusted prompt gains no capability.
3Trust boundaries
Applications, network content, update-hosting infrastructure, prompts, and build workers are untrusted by default. Production signing keys and hardware-backed verified-boot keys are high-trust assets.
4Security review rule
No privileged ObliviOS component enters implementation until its caller, permissions, data minimization, audit event, error behaviour, and test method appear in docs/security/capability-boundaries.md.
Read from docs/security/threat-model.md.