Release 0.3.42 is treated as the fix boundary for CVE-2026-86121, but Imran Siddique’s account says the release changed the server’s default bind address—not its authentication check. That distinction matters: binding to localhost can limit which network interfaces can reach a service, but it does not authenticate requests that still reach it.
What CVE-2026-86121 describes
The vulnerability concerns the Python package cua-computer-server. The GitHub Advisory Database describes versions before 0.3.42 as skipping authentication when CONTAINER_NAME is unset and binding to all interfaces by default. It says an unauthenticated caller could execute shell commands, read and write files, and access interactive PTY shells.
As an Amazon Associate I earn from qualifying purchases.
Those capabilities make the exposure serious when an untrusted party can reach the server. The advisory does not establish that every installation was reachable from the public internet: actual exposure depends on the bind address, network configuration, and any forwarding or proxying in front of the service.
Recommended Free Tools
Why a bind-address change is not an authentication change
A bind address controls where a service listens. Authentication controls whether a request is allowed to use it. These are separate security layers:
#1 Best Overall
- Reachability: A service bound to
0.0.0.0listens on all available IPv4 interfaces. A service bound to127.0.0.1listens on the local loopback interface. - Authorization: An authentication check determines whether a caller may perform an operation after a request reaches the service.
Siddique’s article says version 0.3.42 changed the default from 0.0.0.0 to 127.0.0.1 in the CLI and server constructor, while the relevant authentication code remained unchanged in his comparison. VulnCheck’s reference list also describes a “Bind default changed to 127.0.0.1.” The available advisory records support the binding context, but do not independently establish the full code diff.
Loopback binding can substantially reduce remote network exposure when the service is not otherwise exposed. It is not, by itself, an authentication gate: local processes may still connect, and a forwarded connection may still reach the listener. Whether a particular deployment remains accessible depends on its surrounding network and forwarding configuration.
Rank #2
Why the affected-version records do not line up cleanly
The records identify the same vulnerability context but do not express package scope in the same way. A reader or vulnerability scanner that depends on package-specific affected-version metadata may therefore handle the issue differently from one that recognizes the Git range or interprets the advisory description.
Free tools Windows power users keep installed
One-click scans. No signup required.
| Record | Package and affected range shown | What that means |
|---|---|---|
| GitHub Advisory Database | No package listed; affected-version block is unpopulated. The advisory description says versions before 0.3.42 are affected. | The narrative gives a boundary, but the structured package range is missing in the surfaced record. |
| OSV | A Git range is shown; a package-specific PyPI range is not established in the surfaced record. | Tools consuming the Git range may represent the issue differently from tools expecting a PyPI package range. |
| VulnCheck | cua-computer-server >= 0, < 0.3.42 |
This record explicitly names the package and its affected boundary. |
Siddique also reports that a pip-audit run did not flag the package. That is an account of his run, not an independently reproduced result here; scanner behavior can depend on the advisory data and package metadata available to it.
How to read the severity scores
The scores reported for this CVE use different CVSS versions, so they should not be treated as competing values on one identical scale.
| Score | Source and version |
|---|---|
| 9.8 | CVSS 3.1, reported by Siddique as the NVD score. |
| 9.3 | CVSS 4.0, shown in the surfaced GitHub Advisory Database and OSV records. |
Both figures describe high severity, but the version of CVSS and the source belong with the number whenever it is cited.
Rank #4
What the reported timeline does—and does not—establish
Siddique identifies project issue 1892 as a report dated June 13, 2026, and says it remained open when he checked. The GitHub advisory and OSV record list CVE-2026-86121 as published September 5, 2026; OSV gives a modification date of September 7, 2026. These dates describe the records, not a verified current issue status.
Siddique also identifies version 0.3.46 as published September 10, 2026, and says the allow-all path persisted in that release. That release and source behavior were not independently checked here, so this should be read as his report rather than a verified statement about the latest available code.
Quick Recap
Best Value
Practical implications for users and scanners
- Do not assume the 0.3.42 boundary means an authentication mechanism was added. Siddique’s account identifies a default bind-address change; the records surfaced here do not establish an authentication-code change.
- Assess actual reachability in the deployment, including network interfaces, proxies, tunnels, and forwarded ports. A loopback default is a reduction in exposure, not proof that no remote path exists.
- Do not rely on a single scanner result as proof that the package is unaffected. The surfaced records differ in structured package metadata, and Siddique’s reported
pip-auditrun is not an independently verified universal result. - For a decision about a particular installed release, verify the project’s current release notes and source behavior directly. The available records establish the stated affected boundary and binding reference, but not every release’s complete security behavior.
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
