{"id":"CVE-2026-64651","aliases":["EUVD-2026-46098","GHSA-g48p-5rr5-8rgq"],"status":"published","title":"AI SDK OpenCode Harness Tool Relay Authorization Bypass","description":"The `@ai-sdk/harness-opencode` tool connects HarnessAgent to OpenCode through a sandboxed bridge. Prior to version 1.0.28, the tool relay authorizes requests from any process whose command line contains an allowed helper script path (`host-tool-mcp.mjs`). This allows untrusted code executing in the sandbox to invoke arbitrary host-exposed tools including secret lookups, deployment operations, and cloud API calls without a corresponding model-authorized tool-call event. Exploitation requires a Linux environment (the vulnerable fallback checks `process.platform === 'linux'` and reads `/proc`); an active harness session with one or more host-provided tools; and untrusted code executing in the sandbox (e.g. a malicious dependency, build script, or lifecycle hook) The fix in version 1.0.28 removes the process-path authorization fallback entirely. Relay requests are now only accepted after exact, short-lived, one-time authorization matching the tool name and input from a bridge-observed model event. Some workarounds are available. Do not run the OpenCode harness on untrusted repositories or with untrusted dependencies, and/or limit host-exposed tools to non-sensitive operations when working with untrusted code.","severity":{"version":"4.0","score":6.3,"vector":"CVSS:4.0/AV:L/AC:L/AT:N/PR:N/UI:N/VC:N/VI:L/VA:N/SC:H/SI:H/SA:N","level":"medium","assessor":"cna","from":"cve"},"epss":{"day":"2026-11-12","score":0.00174,"percentile":0.06921,"model":"v2026.06.15"},"facts":{"in_kev":false,"known_exploited":false,"public_exploit":false,"fix_available":false},"cwes":["CWE-863"],"published":"2026-07-20T20:26:07Z","modified":"2026-10-28T02:30:42Z","products":[{"vendor":"vercel","product":"@ai-sdk/harness-opencode"}],"references":[{"url":"https://github.com/vercel/ai/security/advisories/GHSA-g48p-5rr5-8rgq","tags":["x_refsource_CONFIRM","advisory"]},{"url":"https://github.com/vercel/ai/pull/17105","tags":["x_refsource_MISC","fix"]},{"url":"https://github.com/CVEProject/cvelistV5/tree/main/cves/2026/64xxx/CVE-2026-64651.json","tags":["advisory"]},{"url":"https://nvd.nist.gov/vuln/detail/CVE-2026-64651","tags":["advisory"]}],"sources":{"cve":[{"id":"CVE-2026-64651","title":"AI SDK OpenCode Harness Tool Relay Authorization Bypass","description":"The `@ai-sdk/harness-opencode` tool connects HarnessAgent to OpenCode through a sandboxed bridge. Prior to version 1.0.28, the tool relay authorizes requests from any process whose command line contains an allowed helper script path (`host-tool-mcp.mjs`). This allows untrusted code executing in the sandbox to invoke arbitrary host-exposed tools including secret lookups, deployment operations, and cloud API calls without a corresponding model-authorized tool-call event. Exploitation requires a Linux environment (the vulnerable fallback checks `process.platform === 'linux'` and reads `/proc`); an active harness session with one or more host-provided tools; and untrusted code executing in the sandbox (e.g. a malicious dependency, build script, or lifecycle hook) The fix in version 1.0.28 removes the process-path authorization fallback entirely. Relay requests are now only accepted after exact, short-lived, one-time authorization matching the tool name and input from a bridge-observed model event. Some workarounds are available. Do not run the OpenCode harness on untrusted repositories or with untrusted dependencies, and/or limit host-exposed tools to non-sensitive operations when working with untrusted code.","status":"published","published":"2026-07-20T20:26:07.497Z","modified":"2026-07-21T12:49:47.977Z","cvss":[{"version":"4.0","score":6.3,"vector":"CVSS:4.0/AV:L/AC:L/AT:N/PR:N/UI:N/VC:N/VI:L/VA:N/SC:H/SI:H/SA:N","level":"medium","assessor":"cna"}],"cwes":["CWE-863"],"updated_at":"2026-11-02T14:19:58.214427Z"}],"euvd":[{"id":"EUVD-2026-46098","description":"The `@ai-sdk/harness-opencode` tool connects HarnessAgent to OpenCode through a sandboxed bridge. Prior to version 1.0.28, the tool relay authorizes requests from any process whose command line contains an allowed helper script path (`host-tool-mcp.mjs`). This allows untrusted code executing in the sandbox to invoke arbitrary host-exposed tools including secret lookups, deployment operations, and cloud API calls without a corresponding model-authorized tool-call event. Exploitation requires a Linux environment (the vulnerable fallback checks `process.platform === 'linux'` and reads `/proc`); an active harness session with one or more host-provided tools; and untrusted code executing in the sandbox (e.g. a malicious dependency, build script, or lifecycle hook) The fix in version 1.0.28 removes the process-path authorization fallback entirely. Relay requests are now only accepted after exact, short-lived, one-time authorization matching the tool name and input from a bridge-observed model event. Some workarounds are available. Do not run the OpenCode harness on untrusted repositories or with untrusted dependencies, and/or limit host-exposed tools to non-sensitive operations when working with untrusted code.","status":"published","published":"2026-07-20T20:26:07Z","modified":"2026-07-21T12:49:47Z","cvss":[{"version":"4.0","score":6.3,"vector":"CVSS:4.0/AV:L/AC:L/AT:N/PR:N/UI:N/VC:N/VI:L/VA:N/SC:H/SI:H/SA:N","level":"medium"}],"updated_at":"2026-11-02T18:19:47.436242Z"}],"nvd":[{"id":"CVE-2026-64651","description":"The `@ai-sdk/harness-opencode` tool connects HarnessAgent to OpenCode through a sandboxed bridge. Prior to version 1.0.28, the tool relay authorizes requests from any process whose command line contains an allowed helper script path (`host-tool-mcp.mjs`). This allows untrusted code executing in the sandbox to invoke arbitrary host-exposed tools including secret lookups, deployment operations, and cloud API calls without a corresponding model-authorized tool-call event. Exploitation requires a Linux environment (the vulnerable fallback checks `process.platform === 'linux'` and reads `/proc`); an active harness session with one or more host-provided tools; and untrusted code executing in the sandbox (e.g. a malicious dependency, build script, or lifecycle hook) The fix in version 1.0.28 removes the process-path authorization fallback entirely. Relay requests are now only accepted after exact, short-lived, one-time authorization matching the tool name and input from a bridge-observed model event. Some workarounds are available. Do not run the OpenCode harness on untrusted repositories or with untrusted dependencies, and/or limit host-exposed tools to non-sensitive operations when working with untrusted code.","status":"published","published":"2026-07-20T21:16:50.677Z","modified":"2026-10-24T20:10:00.137Z","cvss":[{"version":"4.0","score":6.3,"vector":"CVSS:4.0/AV:L/AC:L/AT:N/PR:N/UI:N/VC:N/VI:L/VA:N/SC:H/SI:H/SA:N/E:X/CR:X/IR:X/AR:X/MAV:X/MAC:X/MAT:X/MPR:X/MUI:X/MVC:X/MVI:X/MVA:X/MSC:X/MSI:X/MSA:X/S:X/AU:X/R:X/V:X/RE:X/U:X","level":"medium","assessor":"secondary"}],"cwes":["CWE-863"],"updated_at":"2026-11-02T14:12:31.024898Z"}],"osv":[{"id":"CVE-2026-64651","title":"AI SDK OpenCode Harness Tool Relay Authorization Bypass","description":"The `@ai-sdk/harness-opencode` tool connects HarnessAgent to OpenCode through a sandboxed bridge. Prior to version 1.0.28, the tool relay authorizes requests from any process whose command line contains an allowed helper script path (`host-tool-mcp.mjs`). This allows untrusted code executing in the sandbox to invoke arbitrary host-exposed tools including secret lookups, deployment operations, and cloud API calls without a corresponding model-authorized tool-call event. Exploitation requires a Linux environment (the vulnerable fallback checks `process.platform === 'linux'` and reads `/proc`); an active harness session with one or more host-provided tools; and untrusted code executing in the sandbox (e.g. a malicious dependency, build script, or lifecycle hook) The fix in version 1.0.28 removes the process-path authorization fallback entirely. Relay requests are now only accepted after exact, short-lived, one-time authorization matching the tool name and input from a bridge-observed model event. Some workarounds are available. Do not run the OpenCode harness on untrusted repositories or with untrusted dependencies, and/or limit host-exposed tools to non-sensitive operations when working with untrusted code.","status":"published","published":"2026-07-20T20:26:07.497Z","modified":"2026-10-28T02:30:42.013125964Z","cvss":[{"version":"4.0","score":6.3,"vector":"CVSS:4.0/AV:L/AC:L/AT:N/PR:N/UI:N/VC:N/VI:L/VA:N/SC:H/SI:H/SA:N","level":"medium"}],"cwes":["CWE-863"],"updated_at":"2026-11-02T15:45:14.096824Z"}]}}