CVE-2026-73079: path traversal leaks pooled AI

CISO Take

Sub2API is an AI API gateway that pools OpenAI/ChatGPT subscription quotas and shares them across tenants, and its POST /responses/*subpath wildcard route fails to validate the client-supplied subpath before splicing it into the upstream request URL. Because the gateway sits between tenants and the operator's shared credentials — ChatGPT/Codex OAuth tokens, OpenAI platform keys, or an operator-configured base URL — any authenticated low-privilege tenant can craft a traversal payload to redirect requests to endpoints the operator never intended to expose, effectively turning the pooled-credential proxy into a confused deputy. There is no public exploit, no Nuclei template, it is not in CISA KEV, EPSS puts exploitation probability at just 0.31%, and CISA's own SSVC decision is TRACK rather than act-now, so this is not an active-exploitation emergency. That said, the attack requires no user interaction, has low complexity (CVSS 3.1: 8.5, AV:N/AC:L/PR:L), and carries a high confidentiality impact, so any org operating or relying on a multi-tenant Sub2API deployment to resell or pool AI provider access should patch to 0.1.169 this week and audit gateway logs for traversal sequences or unexpected upstream hosts in the meantime.

Sources:NVDGitHub AdvisoryEPSSCISA KEVATLAS

What is the risk?

Technically low-complexity to exploit (network, AC:L, no user interaction) but gated behind requiring a valid, even low-privileged, tenant credential — this is not a pre-auth vulnerability. Impact is concentrated in confidentiality (CVSS C:H, I:L, A:N): a malicious or compromised tenant can abuse the operator's pooled provider credentials to reach upstream endpoints outside their authorized scope, including a possible SSRF-style pivot if the operator's configured base URL points at internal infrastructure. Real-world exploitation signals are currently weak — no KEV listing, no public PoC, no scanner template, EPSS 0.31% — and CISA SSVC rates it TRACK, so this sits in the 'patch on normal cadence, don't page anyone' tier for most orgs, escalating only if Sub2API is used to gateway sensitive or high-value pooled accounts.

How does the attack unfold?

Initial Access

Attacker authenticates to Sub2API using their own valid, low-privileged tenant API key.

AML.T0012

Exploitation

Attacker sends a POST to /responses/*subpath with a path traversal payload that Sub2API splices unvalidated into the upstream request URL.

AML.T0049

Credential Relay Abuse

The crafted request is forwarded upstream using the operator's pooled OAuth token or platform API key, reaching an endpoint the tenant was not authorized to target.

AML.T0096

Impact

Attacker gains unauthorized access to upstream data or resources via the pooled credential, with potential SSRF exposure if the operator's base URL routes to internal infrastructure.

Initial Access

Attacker authenticates to Sub2API using their own valid, low-privileged tenant API key.

AML.T0012

Exploitation

Attacker sends a POST to /responses/*subpath with a path traversal payload that Sub2API splices unvalidated into the upstream request URL.

AML.T0049

Credential Relay Abuse

The crafted request is forwarded upstream using the operator's pooled OAuth token or platform API key, reaching an endpoint the tenant was not authorized to target.

AML.T0096

Impact

Attacker gains unauthorized access to upstream data or resources via the pooled credential, with potential SSRF exposure if the operator's base URL routes to internal infrastructure.

How severe is it?

CVSS 3.1

8.5 / 10

EPSS

0.4%

chance of exploitation in 30 days

Higher than 33% of all CVEs

Source: EPSS v3 — FIRST.org

Exploitation Status

No known exploitation

Sophistication

Moderate

What is the attack surface?

AVNetwork

ACLow

PRLow

UINone

SChanged

CHigh

ILow

ANone

What should I do?

1 step

Upgrade Sub2API to 0.1.169 or later immediately — this is the vendor fix and the only complete remediation. Until patched, restrict which subpaths are permitted on the /responses/* route (allowlist known-good upstream paths, reject any path containing '..' or URL-encoded traversal sequences) at a reverse proxy/WAF layer in front of Sub2API. Audit gateway access logs for anomalous subpath values or unexpected upstream hostnames/ports in outbound requests. If the operator's base URL points at any internal service, treat this as a potential SSRF vector and apply network segmentation so the gateway cannot reach internal-only endpoints. Rotate pooled OAuth tokens/API keys if traversal attempts are found in logs, and review tenant-level access scoping to ensure low-privilege accounts cannot reach operator-only routes.

What does CISA's SSVC say?

DecisionTrack

Exploitationnone

AutomatableNo

Technical Impactpartial

Source: CISA Vulnrichment (SSVC v2.0). Decision based on the CISA Coordinator decision tree.

How is it classified?

Auth BypassData LeakageAPIInferenceAML.T0012- Valid Accounts AML.T0049- Exploit Public-Facing Application AML.T0096- AI Service API

Which compliance frameworks are affected?

This CVE is relevant to:

EU AI Act

Article 15-Accuracy, robustness and cybersecurity

ISO 42001

Clause 8.2-AI risk treatment / operational controls

NIST AI RMF

GOVERN 6.1-Policies for third-party and supply chain AI resources

OWASP LLM Top 10

LLM07-Insecure Plugin/Integration Design

Frequently Asked Questions

What is CVE-2026-73079?

Sub2API is an AI API gateway that pools OpenAI/ChatGPT subscription quotas and shares them across tenants, and its POST /responses/*subpath wildcard route fails to validate the client-supplied subpath before splicing it into the upstream request URL. Because the gateway sits between tenants and the operator's shared credentials — ChatGPT/Codex OAuth tokens, OpenAI platform keys, or an operator-configured base URL — any authenticated low-privilege tenant can craft a traversal payload to redirect requests to endpoints the operator never intended to expose, effectively turning the pooled-credential proxy into a confused deputy. There is no public exploit, no Nuclei template, it is not in CISA KEV, EPSS puts exploitation probability at just 0.31%, and CISA's own SSVC decision is TRACK rather than act-now, so this is not an active-exploitation emergency. That said, the attack requires no user interaction, has low complexity (CVSS 3.1: 8.5, AV:N/AC:L/PR:L), and carries a high confidentiality impact, so any org operating or relying on a multi-tenant Sub2API deployment to resell or pool AI provider access should patch to 0.1.169 this week and audit gateway logs for traversal sequences or unexpected upstream hosts in the meantime.

Is CVE-2026-73079 actively exploited?

No confirmed active exploitation of CVE-2026-73079 has been reported, but organizations should still patch proactively.

How to fix CVE-2026-73079?

Upgrade Sub2API to 0.1.169 or later immediately — this is the vendor fix and the only complete remediation. Until patched, restrict which subpaths are permitted on the /responses/* route (allowlist known-good upstream paths, reject any path containing '..' or URL-encoded traversal sequences) at a reverse proxy/WAF layer in front of Sub2API. Audit gateway access logs for anomalous subpath values or unexpected upstream hostnames/ports in outbound requests. If the operator's base URL points at any internal service, treat this as a potential SSRF vector and apply network segmentation so the gateway cannot reach internal-only endpoints. Rotate pooled OAuth tokens/API keys if traversal attempts are found in logs, and review tenant-level access scoping to ensure low-privilege accounts cannot reach operator-only routes.

What systems are affected by CVE-2026-73079?

This vulnerability affects the following AI/ML architecture patterns: API gateways/proxies, multi-tenant LLM platforms, agent frameworks.

What is the CVSS score for CVE-2026-73079?

CVE-2026-73079 has a CVSS v3.1 base score of 8.5 (HIGH). The EPSS exploitation probability is 0.39%.

What is the AI security impact?

Affected AI Architectures

API gateways/proxiesmulti-tenant LLM platformsagent frameworks

MITRE ATLAS Techniques

AML.T0012Valid Accounts

AML.T0049Exploit Public-Facing Application

AML.T0096AI Service API

Compliance Controls Affected

EU AI Act: Article 15

ISO 42001: Clause 8.2

NIST AI RMF: GOVERN 6.1

OWASP LLM Top 10: LLM07

What are the technical details?

Original Advisory

Sub2API is an AI API gateway platform designed to distribute and manage API quotas from AI product subscriptions. From 0.1.135, to 0.1.168, platform API keys issued to tenants are exchanged for upstream requests made with shared provider accounts (ChatGPT/Codex OAuth, OpenAI platform keys, or an operator-configured base URL) that belong to the operator, not to the caller. The `POST /responses/*subpath` wildcard routes spliced the client-supplied subpath into the upstream URL with no validation. This lets an authenticated tenant relay requests to arbitrary upstream endpoints using pooled account credentials via a path traversal. This vulnerability is fixed in 0.1.169.

Exploitation Scenario

An attacker who holds a legitimate, low-privileged Sub2API tenant API key sends a POST request to /responses/<traversal-payload>, where the payload contains path traversal sequences (e.g., encoded '../') instead of a normal Responses API subpath. Sub2API splices this unsanitized subpath directly into the upstream request URL and forwards the request using the operator's shared OAuth token or platform API key — credentials the tenant was never meant to directly control the destination of. The response is relayed back to the attacker, who can use this to probe or reach other upstream paths, potentially other tenants' pooled resources, or — if the operator's base URL is internally routable — internal services reachable from the gateway host, all while the requests appear to originate from the trusted, pooled operator account.

Weaknesses (CWE)

CWE-22Improper Limitation of a Pathname to a Restricted Directory CWE-22Improper Limitation of a Pathname to a Restricted Directory CWE-441Unintended Proxy or Intermediary CWE-441Unintended Proxy or Intermediary

CWE-22 — Improper Limitation of a Pathname to a Restricted Directory: 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.