CVE-2026-77956
CriticalCVSS 10.0Summary
A code injection vulnerability in ash-project ash_ai allows a remote, unauthenticated client to execute arbitrary Elixir code. AshAi.Actions.Prompt evaluates prompt content through EEx.eval_string/2, and when prompt content incorporates request data, attacker-controlled text is compiled and run as an EEx template. This affects ash_ai from 0.1.0 before 1.0.0.
Risk Assessment
An unauthenticated attacker can execute arbitrary code on the server, potentially leading to full system compromise, data theft, or service disruption.
Recommendation
Immediately upgrade ash_ai to version 1.0.0 or later, which includes a fix that stops evaluating function-supplied prompt content as EEx. Until then, restrict access to prompt actions.
Other vulnerabilities in ash_ai
See all- CVE-2026-82580Medium
In ash_ai (from 0.6.0 before 1.0.0), a vulnerability allows disclosure of sensitive information in error messages. Exceptions during tool execution are sent to chat users without filtering, potentially revealing internal technical details.
- CVE-2026-82579Medium
In ash_ai (from 0.6.0 before 1.0.0), a vulnerability allows an infinite loop. An attacker who can influence model outputs can hang the tool loop and cause unbounded, repeated model requests.
- CVE-2026-82564High
An authorization bypass through user-controlled key vulnerability in ash_ai allows a caller to update or destroy records it never identified, including every row in the table. The identity filter is built directly from raw tool arguments.
- CVE-2026-75760High
A vulnerability in ash_ai discloses provider request state and credentials in a user-facing validation error. When the embedding provider call fails, the error contains the raw error term, which may include the URL, response body, and Authorization header with the API key.
- CVE-2026-81315High
An Origin Validation Error vulnerability in ash-project ash_ai allows a malicious web page to bypass the MCP server's DNS-rebinding protection and issue cross-site requests to a user's local MCP server with that user's actor. With default allowed_origins: nil, origin_allowed?/3 accepts an origin when uri.host == conn.host and the forwarded scheme is https, both attacker-controlled. The issue affects ash_ai from 0.8.0 before 1.0.0.
Original NVD description (English source)
Improper Control of Generation of Code (Code Injection) vulnerability in ash-project ash_ai allows a remote, unauthenticated client to execute arbitrary Elixir code. AshAi.Actions.Prompt evaluates prompt content through EEx.eval_string/2. The documented prompt: fn input, context -> ... end form lets the prompt content be built from action arguments, so when a prompt action's text incorporates request data, that attacker-controlled text is compiled and run as an EEx template (Elixir source). Content such as <%= System.cmd(...) %> therefore executes on the server before any model request is made, requiring no authentication beyond reaching a prompt action. The fix stops evaluating function-supplied prompt content as EEx; only statically configured templates are evaluated. This issue affects ash_ai: from 0.1.0 before 1.0.0.