BREACH attack
A side-channel attack that leaks secrets from encrypted HTTPS responses by measuring how well compression matches guessed text.
Presented in 2013, BREACH builds on CRIME from 2012. The attacker makes the victim's browser send requests that include a guess, for example a token that also appears in the response. If the guess is right, the compressed response gets shorter, and the length leaks through the encrypted traffic. With enough variations, the secret can be recovered one character at a time.
What makes BREACH nasty is that it targets HTTP-level compression in the response body, which TLS-level defences do not stop. Mitigations include not reflecting secrets alongside attacker-controlled input, randomising tokens per request, and turning off compression for sensitive pages. Disabling gzip is the quickest fix, and also the most expensive for performance.