Home / Threat model / Attackers
Attackers
Who could attack ObliviOS, the path each takes, and what mitigates, detects and repairs it, with the risk that remains.
1Attacker catalogue
- 1
malicious app
- Path
- Misuses permission, overlay, accessibility, background work, or IPC.
- Mitigation
- Sandbox, scopes, typed policies.
- Detection
- Unexpected resource use, policy denial, anomalous pattern.
- Recovery
- Revoke, restrict, freeze, or remove via typed action.
- Residual risk
- Granted permissions may be abused before detection.
- 2
compromised app update
- Path
- Adds SDK, permission, endpoint, or background behaviour.
- Mitigation
- Difference scanning, signature verification, baseline comparison.
- Detection
- Manifest, SDK, destination, or event change.
- Recovery
- Restrict, isolate, roll back when supported, or remove.
- Residual risk
- Dormant/server-controlled abuse may evade analysis.
- 3
Tracker SDK
- Path
- Exfiltrates identifiers or correlates activity.
- Mitigation
- Per-app networking, classification, profile isolation.
- Detection
- Unexpected destination, volume, or timing.
- Recovery
- Block destination or background traffic.
- Residual risk
- Encryption and domain fronting limit certainty.
- 4
network attacker
- Path
- Observes, manipulates, or blocks traffic.
- Mitigation
- TLS, certificate validation, signed updates.
- Detection
- TLS or update signature failure.
- Recovery
- Reject connection/artifact and use trusted recovery path.
- Residual risk
- Availability can be denied.
- 5
stolen device
- Path
- Attempts offline access or coercive use.
- Mitigation
- Encryption, hardware keys, lockscreen, auto-reboot, profiles.
- Detection
- Failed unlock patterns and configured alerts.
- Recovery
- Lock/reboot per user policy.
- Residual risk
- An unlocked current session remains exposed.
- 6
Co-located adversary with legitimate access
- Path
- Views, or compels an unlock of, the owner's own device (partner, family member, employer) and inspects location-sharing state.
- Mitigation
- Location Anchor's no-persisted-history design: no on/off event is ever written down, so there is nothing to find.
- Detection
- N/A by design — only the current toggle state is observable; there is no history to detect.
- Recovery
- N/A — there is no past record to recover or protect.
- Residual risk
- The status-bar icon (drawn from an interchangeable glyph set) is visible to this adversary if they view the unlocked screen while anchoring is active; the glyph set raises the bar against a casual glance, not against a repeat or attentive observer who already knows it.
- 7
Coercive adversary compelling unlock
- Path
- Physically or legally forces the owner to unlock the device on demand (border search, robbery, an abusive partner) and inspects whatever is shown.
- Mitigation
- Duress Unlock: a second, separately-credentialed, fully isolated decoy profile that a distinct PIN routes directly into at the primary lockscreen, with the real profile force-stopped and never shown.
- Detection
- N/A by design — a decoy unlock is, from every observable signal, a normal unlock.
- Recovery
- N/A — there is no real-profile state exposed to recover from; the owner simply unlocks normally with the real credential at the next opportunity.
- Residual risk
- A sufficiently informed adversary who already suspects ObliviOS ships this feature could demand proof the shown profile is the only one, or examine the OS image offline for the dual-credential-check code path; timing side-channels between a wrong PIN, a decoy-match PIN, and a real-match PIN are a residual risk pending constant-time implementation.
- 8
Physical attacker
- Path
- Uses ports, boot mode, firmware downgrade, or hardware access.
- Mitigation
- Locked bootloader, verified boot, rollback protection, USB policy.
- Detection
- Boot failure or rollback rejection.
- Recovery
- Refuse untrusted boot and restore signed system.
- Residual risk
- Sophisticated hardware attacks can exceed OS controls.
- 9
Compromised build worker
- Path
- Alters source, dependency, or artifact.
- Mitigation
- Pinned manifest, isolated builds, provenance, reproducibility, offline signing.
- Detection
- Hash/provenance mismatch.
- Recovery
- Quarantine, rebuild, rotate exposed credentials.
- Residual risk
- Coordinated source/reviewer/signing compromise is high impact.
- 10
Compromised update server
- Path
- Substitutes, stales, or withholds updates.
- Mitigation
- Signed metadata, rollback protection, mirrors, staged channels.
- Detection
- Signature/version/availability anomaly.
- Recovery
- Reject, pause, publish advisory, use factory image.
- Residual risk
- Availability can be reduced.
- 11
Hostile prompt
- Path
- Tries to induce AI disclosure or action.
- Mitigation
- Minimized inputs, typed allow-list, validation, no shell/root/file authority.
- Detection
- Invalid tool request or denied capability.
- Recovery
- Deny, audit, explain safely.
- Residual risk
- Explanation quality can still be manipulated.
Read from docs/security/threat-model.md.