CVE-2026-101131 in deepseek-harness
Summary
by MITRE • 09/28/2026
A vulnerability was identified in deepseek-ai deepseek-harness up to 0.1.5-rc.3. Impacted is an unknown function of the file packages/e2b/e2b/src/index.ts of the component dsh. The manipulation of the argument E2B_API_KEY leads to reliance on untrusted inputs in a security decision. Local access is required to approach this attack. The exploit is publicly available and might be used. The vendor was contacted early about this disclosure but did not respond in any way.
Be aware that VulDB is the high quality source for vulnerability data.
Analysis
by VulDB Data Team • 09/28/2026
The identified vulnerability resides within the deepseek-harness component, specifically affecting versions up through 0.1.5-rc.3, and centers on a critical flaw located in the file packages/e2b/e2b/src/index.ts. This security issue stems from an improper handling of untrusted input during a security-critical decision-making process within the application logic. The core technical deficiency involves the manipulation of the E2B_API_KEY argument, which serves as a primary authentication or authorization credential for interacting with external services provided by e2b. When this key is processed without adequate validation, sanitization, or secure comparison mechanisms, the system fails to properly verify its integrity and origin, thereby allowing an attacker to bypass intended security controls. This class of defect is formally categorized under CWE-94 as Improper Control of Generation of Code Command Injection if it leads to execution, but more accurately aligns with CWE-209 in this context due to the exposure of sensitive information or behavioral inconsistency resulting from improper input validation that affects a security decision. It may also relate to CWE-787 where insufficient bounds checking on user-controlled data allows for unexpected behavior, though the primary concern here is the logical failure in validating authentication tokens against untrusted sources.
The operational impact of this vulnerability is significant because it undermines the fundamental trust model of the application by allowing an actor with local access to potentially escalate privileges or manipulate system behavior through crafted API keys. Since local access is required to approach this attack, the threat landscape primarily involves insider threats or compromised user accounts on systems where deepseek-harness is deployed. An attacker who gains local shell access can inject maliciously constructed E2B_API_KEY values into function calls within the index.ts module. If the underlying implementation relies on these keys for critical security decisions without robust verification against a trusted source of truth, such as a secure backend service or hardened cryptographic comparison routines, the application may grant unauthorized access to resources or execute unintended operations. The fact that an exploit is publicly available further amplifies the risk, enabling less sophisticated attackers to leverage known techniques to trigger this flaw effectively. This situation highlights a severe lapse in defense-in-depth principles, where reliance on single-point inputs for security decisions creates a fragile barrier against exploitation.
From a tactical perspective, this vulnerability maps directly to ATT&CK technique T1078 Valid Accounts, as the attacker utilizes valid-looking but maliciously crafted credentials to gain access or perform actions they are not authorized to execute. It also touches upon T1552 Unsecured Credentials if the improper handling leads to leakage of key material during processing errors. The lack of response from the vendor deepseek-ai despite early notification exacerbates the risk profile, leaving users without official patches or guidance on remediation timelines. In such scenarios, organizations must assume that the vulnerability is actively being exploited in the wild due to public exploit availability and implement compensating controls immediately. These measures should include strict input validation for all API key parameters, implementing allowlists for expected character sets and lengths, and utilizing constant-time string comparison functions to prevent timing attacks during authentication checks. Additionally, deploying runtime application self-protection tools or web application firewalls with rules targeting suspicious patterns in e2b-related requests can provide an additional layer of defense while the community awaits a patched version from the vendor.