Cryptography Algorithms

Encryption is two-way: with the correct key the original data comes back exactly. Hashing is one-way and keyless, producing a fixed-length fingerprint that cannot be reversed to recover the input. You encrypt data you need to read again, such as a file or a network session. You hash data you only ever need to compare, such as a password or a file integrity check. Encoding, like Base64, is a third thing entirely and offers no protection at all, since anyone can decode it without a secret.

Because security strength measures the work an attacker must do, not the length of the key. Factoring a 2048-bit modulus takes roughly the same effort as brute-forcing a 112-bit symmetric key, thanks to algorithms far better than trial division. Elliptic curve cryptography scales much more efficiently: a 256-bit ECC key delivers 128-bit security, which RSA needs 3072 bits to match. Pushing to 256-bit security would take an RSA key of 15360 bits, and that gap is the reason elliptic curves exist.

Argon2id is the usual first recommendation, since it is memory-hard and therefore resists GPU and ASIC cracking far better than the alternatives. scrypt is a good second choice on the same principle, and bcrypt remains perfectly acceptable provided the work factor is kept current. The exception is a FIPS-regulated environment, where neither Argon2 nor bcrypt is approved and PBKDF2 with a high iteration count is the right answer. The best algorithm and the approved algorithm are not always the same one, which is a distinction worth carrying into any real deployment.

Yes. Kyber was the name of the submission to the NIST post-quantum competition, and ML-KEM, short for Module-Lattice-Based Key-Encapsulation Mechanism, is the name it was given on standardisation as FIPS 203. The same renaming happened to the others: Dilithium became ML-DSA in FIPS 204, SPHINCS+ became SLH-DSA in FIPS 205, and Falcon is becoming FN-DSA. Most search results and a great deal of existing documentation still use the competition names, which is the main reason the mapping is confusing.

To some of it, yes, and the split is what matters. Public-key cryptography is the part genuinely at risk: RSA, Diffie-Hellman and elliptic curves all rest on problems a sufficiently large quantum computer running Shor algorithm could solve, which is why replacements have been standardised as ML-KEM, ML-DSA and SLH-DSA. Symmetric encryption and hashing are far less exposed, since Grover algorithm offers only a quadratic speedup and is awkward to parallelise, so AES-256 and SHA-256 remain sound. No machine capable of any of this exists today, and published estimates put one several orders of magnitude beyond the largest built so far. The reason it is not simply tomorrow problem is harvest now, decrypt later: an attacker can record encrypted traffic today and decrypt it once the capability arrives, so anything that must stay confidential for a decade or more is already exposed.

No, and the asymmetry here is the single most misunderstood point in the topic. Shor algorithm breaks the asymmetric family outright, which means RSA, Diffie-Hellman and elliptic curve cryptography all fail once a sufficiently large quantum computer exists. Grover algorithm only halves the effective strength of symmetric ciphers, so AES-256 retains about 128 bits of security and remains entirely out of reach. The practical consequence is that key exchange and signatures must be replaced, while your bulk encryption largely does not.

It usually means a standards body has withdrawn approval for its use in their scope, which for NIST means United States federal systems. That is a policy decision and not the same as the algorithm being broken. 3DES is the clearest example: it has been withdrawn, yet it has never been defeated and it still functions correctly. MD5 sits at the other end, where the mathematics genuinely failed and no policy is needed to justify avoiding it. This sheet keeps the two ideas apart deliberately, because they lead to different decisions.