praisonaiagents
Affected
- introduced
0, fixed1.6.58(ecosystem)
Fixed versions
1.6.58
Fix
- Fixed version
1.6.58pip install --upgrade praisonaiagents==1.6.58
Advisory
Also known as: CVE-2026-55524, GHSA-vg6p-v9vm-6fgj
praisonaiagents vulnerable to SSRF in web_crawl tool via redirect-following and DNS rebinding (validate-then-fetch gap)
The web_crawl tool performs its SSRF check only on the initial URL: it resolves the hostname once
with socket.gethostbyname and rejects private/loopback/link-local results. It then passes the URL to
a fetcher that uses httpx.Client(follow_redirects=True) - or urllib.request.urlopen when httpx is
absent, which also follows redirects - and re-resolves the hostname at connect time, with no further
validation. This validate-here/fetch-there gap is bypassable two independent ways: HTTP redirects and
DNS rebinding.
Affected code: src/praisonai-agents/praisonaiagents/tools/web_crawl_tools.py
The two bypasses:
Impact:
An agent with web_crawl - driven by direct input or indirect prompt injection - can be made to read
internal-only HTTP services and cloud instance-metadata endpoints (e.g. IAM credentials), with the
response body returned in the tool output. Scope is Changed because the request pivots into the
internal network.
Proof of concept:
A PoC drives the real web_crawl() (httpx absent -> genuine urllib fallback). It runs a loopback
"internal metadata" service and a loopback attacker redirector, substituting DNS only to stand in
for "attacker owns a public domain" / offline routing - the redirect-following and connect-time
re-resolution are the repo's own behavior. Observed:
CONTROL: web_crawl("http://127.0.0.1:.../meta-data/") -> blocked (validator works)
PoC 1A (redirect): attacker.example approved (public); 302 -> loopback metadata
-> result.content leaks {"AccessKeyId":"ASIA_FAKE_STOLEN_CREDENTIAL_..."}
PoC 1B (rebinding): gethostbyname(rebind.example)->public (allowed); connect->127.0.0.1
-> same secret leaked
The control proves the validator blocks a direct loopback request, so the bypasses are genuine.
Remediation:
Resolve the hostname once, validate that IP, and connect to that exact validated IP (pin it) rather
than re-resolving. Disable redirect following (follow_redirects=False; for urllib use a redirect
handler that re-validates), or re-validate every redirect hop's resolved IP. Apply the deny check to
both the validator and the actual socket target. file_tools.py:364 already uses follow_redirects=False
and is the correct pattern to propagate.
Distinct from prior advisories:
The accepted SSRF advisories concern host-string parsing in different code — alternate loopback
encodings in spider_tools (GHSA-5c6w-wwfq-7qqm) and the CLI @url feature (GHSA-5cxw-77wg-jrf3). This
is in the web_crawl tool, which neither advisory names, and the mechanisms (redirect-following and DNS
rebinding) differ categorically from host-string encoding; the spider_tools _host_is_blocked hardening
does not apply to this tool.
High Basis: CVSS v3 score 7.5, vector CVSS:3.1/AV:N/AC:H/PR:N/UI:N/S:C/C:H/I:L/A:N
0, fixed 1.6.58 (ecosystem)1.6.58
1.6.58pip install --upgrade praisonaiagents==1.6.58Fix commands are the export's own generic commands. Ojuju never runs them. Check them against your setup first.
Check your own computer: guard packages check
Source: PyPA Advisory Database · CC BY 4.0