Home / Threat model / Abuse cases
Abuse cases
The ways a privileged path could be abused, the control each one requires and the test evidence for it.
1Privileged-path abuse cases
- 1
AI privilege escalation
- Required control
- Broker denies shell, root, arbitrary Binder/file reads, and direct policy mutation.
- Test evidence
- Typed-tool denial tests.
- 2
Cross-World data leakage
- Required control
- Profile-bound storage/policy and explicit user share contract.
- Test evidence
- Cross-profile isolation and approved-share tests.
- 3
Recovery bypass
- Required control
- Signed factory/recovery artifacts and verified-boot-preserving flow.
- Test evidence
- Tampered-image and interrupted-recovery tests.
- 4
Location Anchor emergency/theft-recovery bypass leakage
- Required control
- Emergency location is checked before the anchor and always wins, unconditionally; the theft-recovery lookup defaults to the real fix and is owner-overridable only through an explicit setting.
- Test evidence
- Simulated-emergency fix-source test; theft-recovery default-path and override-setting tests.
- 5
Location Anchor history persistence
- Required control
- No component may write a timestamped record of Location Anchor's on/off transitions, anchor coordinates, or duration to any store, log, or backup path.
- Test evidence
- Negative test: a full toggle cycle produces zero new persisted artifacts beyond the single current-state record.
- 6
Duress Unlock reveals the real profile under a decoy credential
- Required control
- The keyguard hook must route a decoy-credential match exclusively to the decoy profile; on any hook failure or error path it must fail closed to today's single-credential behavior, never falling through to reveal the real profile.
- Test evidence
- Simulated decoy-vs-real credential collision and hook-failure tests confirming fail-closed-to-normal-path behavior.
- 7
Duress Unlock persists evidence of its own use
- Required control
- No component may write a timestamped record of which credential was entered, when a decoy match occurred, or which profile was last foregrounded, to any store, log, or backup path; a successful decoy match must never count as a failed attempt against the real profile's own lockout counter.
- Test evidence
- Negative test: a full setup-and-decoy-unlock-and-return cycle produces zero new persisted artifacts beyond the pre-existing profile/credential registration.
- 8
Malformed proprietary Bluetooth packet compromises a privileged parser
- Required control
- Memory-safe implementation where feasible, bounded schemas, fuzzing, malformed-input tests, minimal parser authority.
- Test evidence
- AIR-015.
- 9
Stale AirPods location is presented as current
- Required control
- Evidence state and timestamp are mandatory; current state expires immediately when its observation ends.
- Test evidence
- AIR-005, AIR-006.
- 10
One AirPods component's location is assigned to another
- Required control
- Component-specific identity and record; no group inference.
- Test evidence
- AIR-002, AIR-004.
- 11
Malicious app reads AirPods accessory location
- Required control
- Profile-bound private storage and no exported content provider or broad Binder interface.
- Test evidence
- AIR-011.
- 12
Phishing through the optional Apple Find My setup
- Required control
- Exact allow-listed HTTPS origin, browser-owned authentication, visible hostname, no embedded credential surface.
- Test evidence
- AIR-013.
- 13
Apple Find My service outage breaks ordinary AirPods use
- Required control
- The Apple path remains optional and external; local finding and Bluetooth audio remain independent.
- Test evidence
- AIR-014.
- 14
AirPods firmware update silently changes the proprietary protocol
- Required control
- Exact capability registry; changed firmware downgrades to
UNKNOWN; revalidate before restoring the claim. - Test evidence
- AIR-016.
Read from docs/security/threat-model.md.