PyJWT: PyJWKClient still amplifies unauthenticated JWKS fetches on unknown kid values (incomplete fix of CVE-2026-48524)GHSA-2gx3-rcp4-g85q
MEDIUMFix: jpadilla/pyjwt@ba4853aGHSA-2gx3-rcp4-g85q is a medium-severity (CVSS 5.3) CWE-770 vulnerability in pyjwt. A fix is available for pyjwt — see the affected versions and patch details below.
Exploitation Status
Proof-of-concept exploit code exists
- CISA’s SSVC triage found public proof-of-concept exploit code for this CVE, though no confirmed active exploitation.
- CISA assesses this as automatable — exploitation doesn’t require manual, per-target effort, which raises the odds of mass scanning and opportunistic attacks.
Exploitation and automatability from CISA’s SSVC triage for GHSA-2gx3-rcp4-g85q.
EPSS Exploitation Probability
Probability of exploitation in the next 30 days, from FIRST.org EPSS.
How urgent is this, really
Where this sits among everything scored
Real-World Exposure
pyjwtReal-time download stats are indexed for npm and PyPI packages. This vulnerability affects PyPI packages — download data is not available via public APIs for these ecosystems.
Description
Summary CVE-2026-48524 (GHSA-fhv5-28vv-h8m8, "PyJWKClient unbounded JWKS endpoint requests via attacker-controlled kid values (DoS)") was fixed in 2.13.0 by stopping fetch_data() from clearing the cache on a fetch error. That closed one amplification path but did not add the mitigation the advisory's title implies: there is still no rate-limit, negative-cache, or minimum-refresh-interval for unknown kids.
At HEAD, get_signing_key(kid) (jwt/jwks_client.py:185-211), on any unknown kid, calls get_signing_keys(refresh=True), and refresh=True bypasses jwk_set_cache unconditionally and forces a fresh fetch_data(). The kid is read from the unverified token header (get_signing_key_from_jwt decodes with verify_signature=False), so no valid token and no authentication is required. lru_cache does not cache the raised exception, so even the same unknown kid repeated re-fetches on every call.
Affected pyjwt <= 2.13.0 (the latest release; the patched release for CVE-2026-48524). No fixed version yet.
Proof of concept (verified on 2.13.0, cache enabled = realistic prod config) import threading, http.server, socketserver, json from jwt import PyJWKClient hits = {'n': 0} JWKS = json.dumps({"keys":[{"kty":"oct","kid":"real","k":"AAAA"}]}).encode() class H(http.server.BaseHTTPRequestHandler): def do_GET(self): hits['n'] += 1 self.send_response(200); self.send_header('Content-Type','application/json'); self.end_headers() self.wfile.write(JWKS) def log_message(self,*a): pass srv = socketserver.TCPServer(('127.0.0.1',0), H); port = srv.server_address[1] threading.Thread(target=srv.serve_forever, daemon=True).start() c = PyJWKClient(f'http://127.0.0.1/:{port}[/jwks](tg://bot_command?command=jwks).json', cache_keys=True, lifespan=3600) for i in range(8): try: c.get_signing_key(f'attacker-unknown-kid-{i}') except Exception: pass before = hits['n'] for _ in range(5): try: c.get_signing_key('same-unknown') except Exception: pass print('distinct unknown kids: 8 -> fetches:', hits['n']) print('same unknown kid x5 -> extra fetches:', hits['n'] - before)
Output: distinct unknown kids: 8 -> fetches: 9 same unknown kid x5 -> extra fetches: 5 Each unknown kid forces a fresh JWKS fetch; a repeated identical unknown kid still re-fetches every time against an unexpired cache. No rate-limit or negative-cache.
Impact One unauthenticated request -> one outbound JWKS HTTP fetch + full JSON parse on the victim server. An attacker floods tokens carrying junk kids, so the victim hammers its own JWKS/IdP endpoint (amplification: attacker -> victim -> IdP), exhausting victim CPU/sockets and potentially tripping the JWKS provider's rate-limit, causing an application-wide auth outage. This is the unauthenticated DoS the parent advisory is named for, still reachable after the 2.13.0 fix.
Suggested fix Guard the forced refresh on unknown kids: negative-cache unknown kids for a short TTL, or enforce a minimum interval between forced JWKS refreshes, so a repeated or unknown kid cannot force unbounded fetches.
Note: the same-kid-repeated result (5 identical unknown kids producing 5 fetches against an unexpired cache) shows this is request amplification, not legitimate key-rotation handling, since that refresh can never succeed.
Reported by Babakizo (Securva).
Maintainer update — 2026-09-10
The maintainer confirmed the reported behavior against PyJWT 2.13.0. With JWKS caching
enabled, an unknown kid previously forced an unconditional JWKS refresh,
including when the same unknown value was repeated while the cached key set was
still valid. This allowed unauthenticated token headers to cause unnecessary
outbound JWKS requests and repeated parsing work.
The fix is now on master in commit ba4853a. PyJWKClient now applies a
30-second cooldown after successful JWKS fetches before permitting another
unknown-kid refresh, serializes concurrent refresh decisions per client, and
allows callers to configure or disable the cooldown. Cache-disabled behavior
and immediate retry after failed fetches remain unchanged.
Regression tests cover repeated unknown kids, cooldown expiry, concurrent misses, cache-disabled operation, and invalid cooldown values. The available full tox matrix, Ruff, and mypy checks pass. The fix will be included in the next released 2.x version.
Maintainer update — 2026-09-11
The verified fix for this advisory is included in PyJWT 2.14.0, released on 2026-09-11 and available on PyPI. PyJWT 2.14.0 is the first release containing the fix. This advisory is now published with 2.14.0 recorded as the patched version.
Affected Packages
| Ecosystem | Package | Vulnerable range | Fix |
|---|---|---|---|
| 🐍PyPI | pyjwt | all versions | 2.14.0pip install --upgrade 'pyjwt==2.14.0' |
Affected Products
pyjwtpyjwt_projectDetection & mitigation playbook
Open-source dependencyDetect
Scan your dependency tree (package-lock.json, pnpm-lock.yaml, requirements.txt, go.sum, etc.) for pyjwt, including transitive dependencies — a direct dependency you never call can still pull in a vulnerable version.
Fix
Update pyjwt to 2.14.0 or later, then make sure no transitive (indirect) dependency still pins the vulnerable range — O3 confirms GHSA-2gx3-rcp4-g85q is resolved across your whole dependency graph.
Workarounds
Cap what an attacker can consume: apply request size, rate and timeout limits in front of the affected component, and run it with memory and CPU limits so exhaustion degrades one worker rather than the whole service.
Fixing This On Your OS
If you run this on a Linux distribution, patch through your package manager against the distro's own security advisory below — it tracks the exact backported fix for your release, which can ship on a different timeline (and sometimes a different severity) than the upstream project.
| Product | Fixed in | Advisory |
|---|---|---|
| Red Hat AI Inference Server 3.5 | rhaii/vllm-rocm-rhel9:1791425869 | RHSA-2026:82059 |
| Red Hat AI Inference Server 3.5 | rhaii/vllm-gaudi-rhel9:1791427182 | RHSA-2026:82061 |
| Red Hat AI Inference Server 3.5 | rhaii/vllm-cuda-rhel9:1791473631 | RHSA-2026:82368 |
| Red Hat AI Inference Server 3.5 | rhaii/vllm-spyre-rhel9:1791818357 | RHSA-2026:82371 |
| Red Hat AI Inference Server 3.5 | rhaii/vllm-cpu-rhel9:1791473614 | RHSA-2026:82415 |
| Red Hat Enterprise Linux AI 3.5 | rhelai3/disk-image-cuda-rhel9:1793206758 | RHSA-2026:92449 |
| Red Hat Enterprise Linux AI 3.5 | rhelai3/bootc-aws-cuda-rhel9:1793194817 | RHSA-2026:92452 |
Frequently Asked Questions
Is GHSA-2gx3-rcp4-g85q in your dependencies?
Find it across PyPI, including transitive dependencies.