Key Takeaways: Applied Cryptography

A one-page reference. Reread before an exam or before moving on. Dense by design. Builds on Chapter 4's primitives (AES, RSA, ECC, hashing, signatures, PKI, CA, X.509) — review those first if shaky.

The two states of data (the organizing idea)

One datum passes through both states many times; protecting one says nothing about the other. Most real breaches: data encrypted in transit, then written to disk/log/bucket in the clear.

The TLS 1.3 handshake (RFC 8446) — memorize the flow

• 1 round trip (1-RTT) to first app data. (0-RTT resumption exists but is replayable — disable for banking.)
• TLS gives confidentiality (AES/AEAD), integrity (AEAD/MAC), authentication (server cert).
• Attacked at the edges: downgrade (force old proto/cipher), MITM (needs forged/mis-issued cert), expired/bad cert. Defend: disable TLS 1.0/1.1 + SSL, strong cipher policy, HSTS, pinning, monitor negotiated version/cipher.

Reading a cipher suite

TLS 1.2:  TLS_ECDHE_RSA_WITH_AES_128_GCM_SHA256
              │     │        │              └ handshake hash (SHA-256)
              │     │        └ bulk cipher + mode (AES-128-GCM, an AEAD mode)
              │     └ authentication (RSA certificate)
              └ key exchange (ephemeral ECDH → FORWARD SECRECY, the "E")

TLS 1.3:  TLS_AES_128_GCM_SHA256   (only AEAD cipher + hash named; key exchange
                                    & auth negotiated separately, weak options REMOVED)

Forward secrecy = a future theft of the server's long-term key cannot decrypt past recorded sessions. Comes from the ephemeral ("E") key exchange. Defeats harvest-now, decrypt-later.

Cipher/protocol red flags — DISABLE these

Offering a strong cipher does not cancel a weak one — an attacker downgrades to the weak one. Remove the weak option; audit what servers offer, not what your browser negotiates.

VPNs (data in transit, whole-tunnel)

• A VPN protects data in transit between the two endpoints ONLY — not at rest, not past the far end.
• The VPN gateway is a top-tier target: patch it, require MFA, monitor auth, least-privilege access. "On the VPN" ≠ "trusted."

Encryption at rest

Encryption ≠ tokenization. Encryption keeps the data present and reversible wherever the key reaches; tokenization replaces it with a meaningless surrogate, real value confined to a separate vault. For data at rest, the decision is always where the key lives, not which cipher.

Certificate & key lifecycle

generate key → CSR → CA validates → issuance → deploy → MONITOR → RENEW → revoke → retire
                                                          └─ where most failures live ─┘

"When to use what" decision aids

• Protect a web connection → TLS 1.2/1.3, forward-secret AEAD cipher, valid cert, HSTS.
• Protect a whole network link → VPN: IPsec (site-to-site) / WireGuard or TLS-VPN (remote access) + MFA.
• Protect a lost laptop → full-disk encryption (TPM+PIN for high-value/take-home).
• Protect a DB from file theft → TDE; from an insider with query access → column/app-layer encryption.
• Get card data out of scope → tokenization.
• Authenticate a machine/service → mTLS.
• Protect the crown-jewel key itself → HSM.
• Never have a cert expire on you → inventory + expiry monitoring (alert at 30/14/7 days) + automated renewal.

Certification crosswalk

Project additions this chapter

• Meridian program: data-protection standard (in-transit TLS/VPN/mTLS + at-rest FDE/DB encryption/ tokenization/HSM + certificate lifecycle).
• bluekit toolkit: cryptutil.py extended with tls_config_grade(params) → (grade, findings).

Common pitfalls

• "We support strong ciphers" ≠ "we only support strong ciphers" — the weak one must be removed.
• Treating a VPN as protecting data at rest or "everywhere" — it protects the tunnel only.
• Thinking full-disk encryption protects a running machine — it protects a powered-off, lost device.
• TDE on a DB while the master key sits on the same host (no key/data separation).
• Relying on revocation as a safety net (it often fails open) instead of short lifetimes.
• Pinning a certificate and then rotating keys without updating the pins (bricks the app).
• Letting a "convenience"/compatibility fallback downgrade a strong control.
• "Encrypted at rest" passing a compliance checkbox while plaintext leaks to a log/bucket/share.