AI Threat Alert
GHSA-p4pj-vh7h-6cqh: praisonai: unauthenticated path traversal leaks server files
GHSA-p4pj-vh7h-6cqh HIGHPraisonAI's Jobs API accepts arbitrary filesystem paths in the `agent_file` parameter with zero validation and zero authentication, allowing any attacker with network access to port 8005 to read any file the server process can access. For AI agent deployments — where `.env` files and `/proc/1/environ` routinely contain LLM API keys, database credentials, and SSH private keys — a single curl command is enough to exfiltrate the full credential set needed to pivot into cloud infrastructure and AI service accounts. The public proof-of-concept requires no expertise, and with no EPSS or KEV signals yet available, opportunistic scanning is the primary near-term threat vector. Upgrade to praisonai >= 4.6.59 immediately; if patching is blocked, firewall port 8005 to trusted CIDRs only and rotate all secrets stored in server environment files.
What is the risk?
High risk with elevated urgency for exposed deployments. CVSS 7.5 accurately reflects a network-accessible, zero-authentication, low-complexity attack with high confidentiality impact. For AI environments the practical risk exceeds the base score: AI agent servers are a concentration point for LLM API keys, cloud credentials, and inter-service tokens. A single exploitation event can yield key material sufficient to impersonate the victim's AI workloads, drain API budgets, or pivot laterally into MLOps infrastructure. The trivial PoC lowers the bar to near-zero for opportunistic attackers scanning for exposed agent frameworks on non-standard ports.
How does the attack unfold?
What systems are affected?
| Package | Ecosystem | Vulnerable Range | Patched |
|---|---|---|---|
| PraisonAI | pip | < 4.6.59 | 4.6.59 |
Do you use PraisonAI? You're affected.
How severe is it?
What is the attack surface?
What should I do?
6 steps-
Patch immediately: upgrade praisonai to >= 4.6.59 which introduces path validation on the
agent_filefield. -
If patching is blocked: restrict network access to port 8005 to trusted CIDRs only via firewall or reverse proxy.
-
Rotate all secrets: any API keys, tokens, or private keys accessible to the server process must be considered compromised if the endpoint was internet-exposed.
-
Detect exploitation: search access logs for POST requests to
/api/v1/runscontainingagent_filevalues with absolute paths (starting with/) or path traversal sequences (../). -
Audit exposure: verify whether port 8005 is internet-facing across all environments (dev, staging, prod) — developer environments frequently expose this accidentally.
-
Apply network segmentation: AI agent infrastructure should not be directly internet-reachable; place behind an authenticated reverse proxy.
How is it classified?
Which compliance frameworks are affected?
This CVE is relevant to:
Frequently Asked Questions
What is GHSA-p4pj-vh7h-6cqh?
PraisonAI's Jobs API accepts arbitrary filesystem paths in the `agent_file` parameter with zero validation and zero authentication, allowing any attacker with network access to port 8005 to read any file the server process can access. For AI agent deployments — where `.env` files and `/proc/1/environ` routinely contain LLM API keys, database credentials, and SSH private keys — a single curl command is enough to exfiltrate the full credential set needed to pivot into cloud infrastructure and AI service accounts. The public proof-of-concept requires no expertise, and with no EPSS or KEV signals yet available, opportunistic scanning is the primary near-term threat vector. Upgrade to praisonai >= 4.6.59 immediately; if patching is blocked, firewall port 8005 to trusted CIDRs only and rotate all secrets stored in server environment files.
Is GHSA-p4pj-vh7h-6cqh actively exploited?
No confirmed active exploitation of GHSA-p4pj-vh7h-6cqh has been reported, but organizations should still patch proactively.
How to fix GHSA-p4pj-vh7h-6cqh?
1. Patch immediately: upgrade praisonai to >= 4.6.59 which introduces path validation on the `agent_file` field. 2. If patching is blocked: restrict network access to port 8005 to trusted CIDRs only via firewall or reverse proxy. 3. Rotate all secrets: any API keys, tokens, or private keys accessible to the server process must be considered compromised if the endpoint was internet-exposed. 4. Detect exploitation: search access logs for POST requests to `/api/v1/runs` containing `agent_file` values with absolute paths (starting with `/`) or path traversal sequences (`../`). 5. Audit exposure: verify whether port 8005 is internet-facing across all environments (dev, staging, prod) — developer environments frequently expose this accidentally. 6. Apply network segmentation: AI agent infrastructure should not be directly internet-reachable; place behind an authenticated reverse proxy.
What systems are affected by GHSA-p4pj-vh7h-6cqh?
This vulnerability affects the following AI/ML architecture patterns: agent frameworks, multi-agent orchestration, model serving, AI development environments, cloud-hosted AI pipelines.
What is the CVSS score for GHSA-p4pj-vh7h-6cqh?
GHSA-p4pj-vh7h-6cqh has a CVSS v3.1 base score of 7.5 (HIGH).
What is the AI security impact?
Affected AI Architectures
MITRE ATLAS Techniques
AML.T0025 Exfiltration via Cyber Means AML.T0037 Data from Local System AML.T0049 Exploit Public-Facing Application AML.T0055 Unsecured Credentials AML.T0083 Credentials from AI Agent Configuration Compliance Controls Affected
What are the technical details?
Original Advisory
### Summary An unauthenticated attacker can read arbitrary files on the server by supplying an absolute filesystem path in the `agent_file` field of the Jobs API. The field has no path validation, no allowlist, and no authentication is required to submit jobs. ### Details The `agent_file` field in `JobSubmitRequest` accepts any filesystem path with no validation: ```python # src/praisonai/praisonai/jobs/models.py:29 agent_file: Optional[str] = Field(None, description="Path to agents.yaml file") # NO path validator, NO allowlist ``` The executor reads the file directly: ```python # src/praisonai/praisonai/jobs/executor.py:221 agent_file = job.agent_file or "agents.yaml" # passed directly to yaml.safe_load(open(agent_file)) ``` ### Proof of Concept ```bash curl -X POST http://:8005/api/v1/runs \ -H "Content-Type: application/json" \ -d '{"prompt": "run", "agent_file": "/etc/passwd"}' ``` Server responds with contents of `/etc/passwd`. Other exploitable paths: - `/proc/1/environ` — environment variables, API keys - `/home//.ssh/id_rsa` — SSH private keys - `/app/.env` — application secrets ### Impact Any unauthenticated attacker with network access to port 8005 can read any file accessible to the server process, including credentials, private keys, and environment variables.
Exploitation Scenario
An attacker scans the internet for services on port 8005 and discovers an exposed PraisonAI instance running in a cloud environment. Without any credentials, they issue a POST to `/api/v1/runs` with `agent_file: /proc/1/environ`, receiving back the process environment containing `OPENAI_API_KEY`, `ANTHROPIC_API_KEY`, `DATABASE_URL`, and `AWS_SECRET_ACCESS_KEY`. They repeat with `/app/.env` and `/home/ubuntu/.ssh/id_rsa` to collect the full credential set. Within minutes the attacker holds API keys to impersonate the victim's AI workloads, an SSH key for direct host access, and cloud credentials to enumerate and exfiltrate connected storage — all without any prior authentication or AI/ML knowledge.
Weaknesses (CWE)
CWE-22 — Improper Limitation of a Pathname to a Restricted Directory ('Path Traversal'): The product uses external input to construct a pathname that is intended to identify a file or directory that is located underneath a restricted parent directory, but the product does not properly neutralize special elements within the pathname that can cause the pathname to resolve to a location that is outside of the restricted directory.
- [Implementation] Assume all input is malicious. Use an "accept known good" input validation strategy, i.e., use a list of acceptable inputs that strictly conform to specifications. Reject any input that does not strictly conform to specifications, or transform it into something that does. When performing input validation, consider all potentially relevant properties, including length, type of input, the full range of acceptable values, missing or extra inputs, syntax, consistency across related fields, and conformance to business rules. As an example of business rule logic, "boat" may be syntactically valid because it only contains alphanumeric characters, but it is not valid if the input is only expected to contain colors such as "red" or "blue." Do not rely exclusively on looking for malicious or malformed inputs. This is likely to miss at least one undesirable input, especially if the code's environment changes. This can give attackers enough room to bypass the intended validation. However, denylis
- [Architecture and Design] For any security checks that are performed on the client side, ensure that these checks are duplicated on the server side, in order to avoid CWE-602. Attackers can bypass the client-side checks by modifying values after the checks have been performed, or by changing the client to remove the client-side checks entirely. Then, these modified values would be submitted to the server.
Source: MITRE CWE corpus.
CVSS Vector
CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:H/I:N/A:N References
Timeline
Related Vulnerabilities
CVE-2026-61447 10.0 PraisonAI: RCE via unsandboxed LLM code execution
Same package: praisonai CVE-2026-61445 9.9 PraisonAI: AICoder root RCE via unsanitized tool calls
Same package: praisonai CVE-2026-47392 9.9 praisonaiagents: RCE via Python sandbox bypass
Same package: praisonai GHSA-vmmj-pfw7-fjwp 9.9 praisonai: sandbox escape gives RCE via codeMode tool
Same package: praisonai GHSA-vc46-vw85-3wvm 9.8 PraisonAI: RCE via malicious workflow YAML execution
Same package: praisonai