Authlib CVE-2025-62706

Exploitation requires that the application uses Authlib's JOSE JWE decode path and accepts/decrypts JWE tokens whose protected header specifies 'zip':'DEF' (DEFLATE compression), on versions prior to 1.6.5. The attacker must be able to supply a token the target will actually decrypt - i.e. possess or control the relevant decryption key material or operate within a trust relationship that accepts externally-supplied JWEs (reflected by PR:L). Limiting factors: services that never enable/accept zip=DEF, that strip or reject compressed JWEs at the boundary, that enforce strict maximum token sizes and rate limiting, or that already run 1.6.5+ are not exploitable. No end-user interaction is required (UI:N). Impact is availability-only (memory/CPU exhaustion); no data disclosure or modification occurs.

This is a genuine but moderate-priority availability issue rather than a critical emergency. The CVSS vector (AV:N/AC:L/PR:L/UI:N/S:U/C:N/I:N/A:H, base 6.5) correctly captures a network-reachable, low-complexity attack with no confidentiality or integrity impact and high availability impact - a classic CWE-400 resource-exhaustion (decompression bomb) profile. Multiple signals align and should temper urgency: EPSS is only 0.46% (38th percentile), there is NO CISA KEV listing (no evidence of active exploitation), and the only impact is denial of service. These point to low real-world mass-exploitation likelihood. Countervailing the low EPSS, publicly available exploit/PoC code exists (the vendor GHSA ships a working jwe_deflate_dos_demo.py), which lowers the skill barrier and makes opportunistic DoS against exposed Authlib JOSE endpoints plausible. One important nuance that limits risk: the PR:L in the vector and the description ('an attacker who can supply decryptable tokens') mean the attacker must be able to produce JWEs the target will actually decrypt - i.e. possess or control the relevant key material or a trust relationship - so this is not a trivially anonymous internet-wide attack. SSVC-style reading: exploitation=PoC, automatable=likely yes (repeatable crafted tokens), technical impact=partial (availability only). Net: patch in due course, prioritize internet-facing services that accept third-party or externally-issued JWE tokens with zip=DEF, but this does not rise to drop-everything severity.

An attacker who can submit a JWE the service will decrypt crafts a tiny zip=DEF (DEFLATE) ciphertext containing highly repetitive data that inflates to tens or hundreds of megabytes on decompression. When Authlib's JWE decode path decompresses it without a size bound, the target process spikes RSS and CPU; repeating the request across connections starves the host and causes denial of service. Public PoC code (jwe_deflate_dos_demo.py in the vendor advisory GHSA-g7f3-828f-7h7m) demonstrates the amplification, lowering the barrier to reproduce.