Threat Model

What Ephemera protects against, what it does not, and why.

Threats & Mitigations

Threat Mitigation
Stolen SSH private key Keys are ephemeral. Certificates expire in minutes, limiting exposure window.
Credential phishing WebAuthn is phishing-resistant. TOTP is fallback only.
Unauthorized privilege escalation JIT Sudo requires fresh MFA approval per session. Logged centrally.
Audit log tampering Merkle-chained ledger. Remote syslog mirroring to isolated sink.
CA key compromise Encrypted at rest. Zero-downtime rotation with multi-CA trust model.
Single-point recovery failure Shamir secret sharing (k-of-n) for backup decryption.

Explicit Non-Goals

The following are intentionally excluded from Ephemera's scope:

Feature Reason
SSH MITM Proxy Violates end-to-end encryption; adds attack surface.
Custom SSH Protocol Unnecessary complexity. Native OpenSSH is sufficient.
Cloud Dependency Designed for sovereign, air-gapped deployments.
Session Replay UI Recording research is decentralized; no central player planned.
QR-Based Shard Recovery JSON file shards are simpler and more portable.
C PAM Module pam_exec with Python is portable and auditable.

Inherent Security Trade-offs

The following behaviors are inherent properties of Ephemera's technical design (time-bound credentials, availability models, and quorum cryptography) and are documented as expected boundaries of the threat model:

Security Disclosure Philosophy

Certain implementation details are intentionally omitted from public documentation. Ephemera follows a "disclose-the-model, protect-the-keys" philosophy.

Commitment to Transparent Security
Explicit acknowledgment of security boundaries increases systemic trust. In high-assurance infrastructure, documenting the technical limits of cryptographic models is essential for informed risk management.