• cost is paid even when signature is invalid
• the decoded bytes are discarded
• attacker controls the size of this cost via token length

Impact (evidence-driven)

Security impact

• Unauthenticated remote DoS: decoding work happens before signature rejection → attacker does not need signing key.
• CPU amplification: Base64URL decode time scales linearly with payload segment size.
• Memory amplification: decoded output allocates large byte buffers (tens of MB per request).
• Operational impact: request queueing / worker starvation under modest concurrency bursts.

Standards context (RFC 7797)

RFC 7797 explicitly notes this option is used when payload is large and/or detached, and discusses interoperability requirements around marking it critical (“crit” with “b64”). ([IETF Datatracker][2]) (PyJWT supports crit validation, but the issue here is decode order / unbounded decode of an unused segment.)

Affected Versions

• Confirmed affected: PyJWT 2.12.1 (tested from your local editable install and repo).
• Likely affected: all versions that include detached payload support for JWS decoding, which was introduced in 2.4.0 (“Add detached payload support for JWS encoding and decoding”). ([pyjwt.readthedocs.io][3])

(For GHSA, this phrasing is strong: “confirmed” + “likely since feature introduction”.)

Threat Model

Typical real deployment

A service verifies signed HTTP requests or webhooks using detached JWS:

• token is provided in JSON body / query / header
• actual payload is the HTTP request body passed as detached_payload

Attacker

• remote unauthenticated client
• can send requests to verify endpoint
• does not need a valid signature (invalid signature still triggers the expensive decode path)

Attack chain

1. Attacker crafts a JWS compact token with header containing "b64": false and crit:["b64"].
2. Attacker inflates the payload segment (middle segment) to millions of Base64URL characters.
3. Server calls PyJWS.decode(...detached_payload=...).
4. PyJWT decodes the inflated segment (CPU + memory).
5. Signature is rejected afterward (401) — but resources already consumed.
6. Repeated requests or bursts cause queueing/worker starvation → DoS.

Proof of Concept - file names + results

PoC placement

• server_localhost.py

• client_localhost.py

• flood_localhost.py

PoC # 1 - Localhost verification server

File: server_localhost.py

Purpose: real HTTP endpoint (POST /verify) that calls PyJWT detached verification and prints: ok / time_ms / peak_bytes / token_len / error.

Results (server console output)

[+] Listening on http://127.0.0.1:8000
[+] POST /verify  JSON: {"token": "..."}

[127.0.0.1] ok=True  time_ms=0.102 peak_bytes=2624     token_len=117      err=None
[127.0.0.1] ok=False time_ms=2.012 peak_bytes=2000983  token_len=500078   err=InvalidSignatureError
[127.0.0.1] ok=True  time_ms=1.591 peak_bytes=2001061  token_len=500117   err=None

[127.0.0.1] ok=True  time_ms=0.065 peak_bytes=2304     token_len=117      err=None
[127.0.0.1] ok=False time_ms=7.534 peak_bytes=8000983  token_len=2000078  err=InvalidSignatureError
[127.0.0.1] ok=True  time_ms=6.347 peak_bytes=8001061  token_len=2000117  err=None

[127.0.0.1] ok=True  time_ms=0.066 peak_bytes=2304     token_len=117      err=None
[127.0.0.1] ok=False time_ms=23.034 peak_bytes=32000983 token_len=8000078 err=InvalidSignatureError
[127.0.0.1] ok=True  time_ms=22.097 peak_bytes=32001061 token_len=8000117 err=None

Key takeaways from these results

• At 8,000,000 chars, a single invalid-signature request still causes:

  • ~23 ms server work
  • ~32 MB peak allocations
  • returns 401 (invalid signature) → attacker does not need key.

PoC # 2 - Localhost network client

File: client_localhost.py Purpose: generates baseline + (invalid signature) + (valid signature) tokens and sends them over HTTP to localhost server.

Results (client output)

payload-chars = 500,000

=== BASELINE (valid b64=false token) ===
HTTP: 200
client_wall_ms: 6.3499...
server_time_ms: 0.10197...
server_peak_bytes: 2624

=== ATTACK (INVALID signature - attacker needs no key) ===
HTTP: 401
client_wall_ms: 4.1010...
server_time_ms: 2.01217...
server_peak_bytes: 2000983
error: InvalidSignatureError

=== ATTACK (VALID signature - accepted path still wastes) ===
HTTP: 200
client_wall_ms: 3.6586...
server_time_ms: 1.59092...
server_peak_bytes: 2001061

payload-chars = 2,000,000

=== BASELINE ===
HTTP: 200
server_time_ms: 0.06527...
server_peak_bytes: 2304

=== ATTACK (INVALID signature) ===
HTTP: 401
server_time_ms: 7.53430...
server_peak_bytes: 8000983

=== ATTACK (VALID signature) ===
HTTP: 200
server_time_ms: 6.34682...
server_peak_bytes: 8001061

payload-chars = 8,000,000

=== BASELINE ===
HTTP: 200
server_time_ms: 0.06573...
server_peak_bytes: 2304

=== ATTACK (INVALID signature) ===
HTTP: 401
server_time_ms: 23.03403...
server_peak_bytes: 32000983

=== ATTACK (VALID signature) ===
HTTP: 200
server_time_ms: 22.09702...
server_peak_bytes: 32001061

Why this is strong evidence

• The server clearly does heavy work before rejecting invalid signatures.
• The “valid signature” case shows even accepted requests waste resources due to unused payload segment.

PoC # 3 - Localhost flood / burst concurrency

File: flood_localhost.py Purpose: sends N concurrent invalid-signature requests over HTTP to demonstrate queueing/worker starvation.

Results (your run: 20 concurrent @ 8,000,000 chars)

total_wall_ms: 1374.5405770000616

(16, 401, 1156.4504789998864, 21.350951999920653, 32000983, 'InvalidSignatureError')
(19, 401, 1151.2852699997893, 21.208721999755653, 32000983, 'InvalidSignatureError')
(18, 401, 1102.7211239997996, 21.685218999664357, 32000983, 'InvalidSignatureError')
(13, 401, 1102.0718189997751, 21.26572200040755, 32000983, 'InvalidSignatureError')
(11, 401, 1095.9345460000804, 20.586017000368884, 32000983, 'InvalidSignatureError')
(17, 401, 1085.2552810001725, 22.893039000337012, 32000983, 'InvalidSignatureError')
(10, 401, 1078.3629560000918, 22.737160999895423, 32000983, 'InvalidSignatureError')
(7,  401, 1048.2011740000416, 22.476282000297942, 32000983, 'InvalidSignatureError')
(8,  401, 378.93017700025666, 21.377330999712285, 32000983, 'InvalidSignatureError')
(1,  401, 281.45106800002395, 21.34223099983501, 32000983, 'InvalidSignatureError')

Interpretation

• Each request still costs ~20–23 ms server processing and ~32 MB peak allocations.
• But client-observed latency rises up to ~1.15 seconds because requests queue behind each other → clear worker starvation/HoL blocking.
• All were rejected with 401 InvalidSignatureError → still unauthenticated.

Fix

Goal

Prevent unbounded resource consumption from an attacker-controlled payload segment that is unused in b64=false detached flow.

Minimal change strategy

In _load() (or by refactoring parse order), do not Base64-decode payload_segment until after you know whether b64=false applies.

Two safe options:

1. Reject non-empty payload segment when b64=false

   • Parse header first
   • If b64 is false and payload_segment is non-empty → raise DecodeError before decoding
   • Then verification uses detached_payload only

2. Skip decoding payload segment entirely when b64=false

   • Keep payload segment as raw bytes or empty
   • Use detached payload for signing input

This aligns with the idea that detached payload is the trusted payload input for verification; the compact payload segment should not become a resource amplification vector.

(Implementation context: the current decode order and unconditional base64url_decode(payload_segment) are visible in the file and line region around _load() and decode_complete() ([GitHub][1]).)

Workarounds

• Enforce strict max token length at the HTTP boundary (proxy/gateway).
• Apply rate limiting on verification endpoints.
• If detached JWS (b64=false) is not needed in your app, reject tokens where header includes "b64": false.