New features
- New module
Crypto.PublicKey.SLHDSA, for SLH-DSA keys (FIPS 205), with all
twelve parameter sets (SHA2 and SHAKE, categories 1, 3 and 5, small or fast).
The algorithms of FIPS 205 are in Python; the functions that compute most of the hashes
(WOTS+ chains, XMSS and FORS tree nodes) are in C, on top of the SHA-2 and Keccak code
of PyCryptodome.
The C functions compute several hashes at once where the CPU allows it:
4 SHA-256 hashes with the Intel SHA extensions (interleaved) or the ARMv8 SHA-256
instructions, 8 SHA-256 hashes with AVX2 (when there are no SHA extensions),
and 4 SHAKE256 hashes with AVX2 and BMI2 (4-way Keccak).
The WOTS+ chains of a key pair are computed in one call to C. - SLH-DSA keys can be exported and imported as defined in RFC 9909 (pure SLH-DSA):
public keys as X.509SubjectPublicKeyInfo(also from certificates),
private keys as PKCS#8OneAsymmetricKey(also encrypted), in DER or PEM. - New module
Crypto.Signature.slhdsa, for SLH-DSA and HashSLH-DSA signatures (FIPS 205),
hedged (default) or deterministic, with an optional context. - SLH-DSA signing, key generation and
construct()(which checks the private key)
accept a parameterthreads, to compute the XMSS and FORS trees on several threads.
With 8 threads, signing with thesparameter sets is 3.5x to 5x faster. Crypto.PublicKey.SLHDSAalso supports the six parameter sets of the initial public
draft of NIST SP 800-230 (April 2026), such asSLH_DSA_SHA2_128_24: smaller signatures
and faster verification, but each key must sign at most 2^24 messages, and key generation
and signing take minutes (usethreads). The draft may still change: their support is
experimental, and they must not be used in production.- New self-test option
--include-very-slow-tests, for the tests markedvery_slow
(like key generation and signing with the SP 800-230 parameter sets, which take about
half an hour): they do not run otherwise. - New module
Crypto.Cipher.AEGIS, for the authenticated ciphers AEGIS-128L and
AEGIS-256 (RFC 10032), with 16- or 32-byte tags. They use the AES instructions of the CPU
(AES-NI or the ARMv8 Cryptography Extensions): AEGIS-128L is then 3 to 5 times faster
than AES-GCM. Without them, the AES round is bitsliced (constant time), with NEON
on ARM and SSE2 on x86; on 64-bit ARM, AEGIS-128L is then still faster than
ChaCha20-Poly1305.
Breaking changes
- In
Crypto.Signature.mldsa.new(), the context can only be passed as a keyword
argument (context=), as forCrypto.Signature.slhdsa.new().
Resolved issues
- Security, GH#968:
ECC.construct(), and soECC.import_key(), accepted the point at infinity
as a public key on the NIST P curves and secp256k1 (for instance,04followed by zero bytes
in aSubjectPublicKeyInfo). ECDSA signatures for such a key can be forged without
any secret. It is now rejected withValueError. Thanks to Zulfareen Shaikh. - Security, GH#969:
DSA.construct(), and soDSA.import_key(), accepted public keys
equal to 1 or outside the subgroup of orderq(for instance,p - 1).
Fory = 1, a signature can be forged on any message without any secret.
They are now rejected withValueError(NIST SP 800-89, 5.3.1). Thanks to Zulfareen Shaikh. - GH#966: in the
decrypt_and_verify()method of GCM, CCM, EAX and SIV, theoutputbuffer
is filled with zeroes if the MAC tag is not valid, so that no unauthenticated plaintext
is left in it (as GCM-SIV already did). Thanks to Zulfareen Shaikh. - Security: decoding a DER
INTEGERtook time quadratic in its length, so a crafted
key of a few hundred KB (for instance, forRSA.import_key) kept the CPU busy for minutes.
It now takes linear time. Decoding anOBJECT IDENTIFIERhad the same problem: its
subidentifiers are now limited to 128 bytes. Thanks to netherportal3. - GH#965: DSA key import now rejects non-integer domain parameters with
ValueError. Thanks to Mrmaxmeier. - GH#963: Fixed the build on the x32 ABI (ILP32 on x86-64). Thanks to Kenny Cheng.
- GH#864: The type hints of the scalar multiplication of EC points accept
Integerobjects.
Thanks to danrr and hakarrr.