The NIP-04 decryption parser split attacker-controlled content on every ?iv= separator and collected all resulting segments before checking that the message had the expected two parts. It also Base64-decoded the complete IV text before checking that it represented the required 16-byte AES-CBC IV.

A malicious sender could include a large number of separators or an oversized IV in an encrypted direct message. Applications that attempted to decrypt the message performed avoidable allocations proportional to the malformed input, with additional allocation amplification from the segment vector and Base64 output. This can consume memory and CPU in clients processing messages received through a relay. It does not weaken NIP-04 encryption or reveal plaintext or key material.

The parser now uses a single bounded split, rejects additional separators, and validates the 24-byte encoded IV length before Base64 decoding. Malformed inputs are returned as errors without allocating for every separator or decoding an arbitrarily large IV.