A risk model for CVE-2026-114125: why a conditional chain still scores 9.4
A high severity score can look inconsistent with an advisory that describes several conditions and no known exploitation. CVE-2026-114125 is a useful example of why the two statements can both be true. CVSS measures the potential impact and exploitability of a vulnerability under its stated metrics. Exploitation observations answer a different question.
What the CVSS v4 score says
The GitHub Reviewed advisory assigns 9.4 Critical using CVSS:4.0/AV:N/AC:L/AT:N/PR:N/UI:P/VC:H/VI:H/VA:H/SC:H/SI:H/SA:H. The attack vector is Network, attack complexity is Low, attack requirements are None, and no privileges are required. Passive user interaction is present. The vulnerable system and subsequent system impacts are high for confidentiality, integrity, and availability.
That vector reflects a potentially serious outcome. A malicious website can provide the network-facing content while the developer's browser supplies access to a local Storybook server. No prior account on the target is required. If the chain succeeds, the developer's project and permissions become the relevant boundary.
The advisory also says the attack relies on correctly exploiting a sequence of many vulnerabilities, some of which were not disclosed to the Storybook team. That wording limits what can be inferred from the score. CVSS is not a claim that the published advisory contains a simple one-request exploit.
Likelihood has separate evidence
The GitHub advisory lists an EPSS score of 0.385 percent, in the 33rd percentile. Storybook says it is not aware of exploitation in the wild. Those signals suggest a lower observed or predicted urgency than the critical rating alone might imply. They do not establish that no attacker will use the chain, and EPSS is a time-bounded estimate rather than a verdict on one organization's exposure.
Risk assessment should combine the score with local conditions. A vulnerable dev server that a person opens while browsing untrusted sites is materially different from a server that is not installed or never runs. A workstation with production credentials and broad network access has a different blast radius from an isolated test environment.
Why file modification matters
The advisory's worst case starts with unauthorized changes to story files. Story files are inputs to development and testing workflows, not inert documentation. They may be imported by a test runner, browser runtime, or local tool. If the attacker can shape those files and cause a later process to execute them, the impact expands from repository integrity to code execution.
Several outcomes deserve separate treatment:
- A malicious site reaches the preview but no file changes are observed.
- Story files change, but no execution path consumes the modification.
- Changed files enter a local process with the developer's privileges.
- The process can access credentials, source, or internal systems beyond the repository.
Only the stronger outcomes justify claims of code execution or credential compromise. The lower outcomes still require investigation and remediation because they show that the trust boundary failed.
The vulnerable population is not one number
The affected package ranges cover Storybook 8, 9, 10, and 11 alpha releases. Built output is not affected, and dev servers that are never opened in a browser fall outside the published path. A ZoomEye search for title="Storybook" produced 5,438 matches, while the CVE-specific filter produced zero. The title count is a product fingerprint, not a vulnerable-version census, and it cannot measure private developer laptops.
Organizations should score exposure at the asset level. Inputs include installed version, interactive browser use, bind address, browser localhost permissions, feature flags, and the privileges of local test and build processes. This produces a more useful likelihood estimate than applying 9.4 uniformly to every Storybook title.
Turning the model into action
Patch to 8.6.19, 9.1.21, 10.6.2, or 11.0.0-alpha.6. Stop affected interactive servers until the update is in place. Remove insecure iframe and messaging flags unless a reviewed compatibility requirement remains. Keep development servers on loopback where possible and preserve host validation when LAN sharing is necessary.
The patched defaults are part of risk reduction. They block cross-host development iframe embedding and reject unknown message origins. Update Chromatic's Visual Tests addon to 5.5.1 or later where needed. Then verify the runtime, not just the package declaration.
A conditional exploit can still justify a critical score when successful execution reaches developer permissions. The practical response is to reduce the number of environments that meet those conditions and to preserve evidence when the chain may have progressed.
Top comments (0)