ash_ai Vulnerabilities Expose Critical Risks in Elixir Framework

On August 30, 2026, maintainer zachdaniel disclosed a cluster of six vulnerabilities in ash_ai, an LLM toolbox extension for the Elixir-based Ash Framework. This marks the first coordinated security disclosure in the Elixir/Ash ecosystem. The most critical vulnerability, CVE-2026-77956 (CVSS 8.9), allows remote code execution via user-provided prompts treated as executable code. Other vulnerabilities include CVE-2026-81315 (CVSS 7.4), which enables origin validation bypass, and CVE-2026-82564 (CVSS 7.1), allowing authorization bypass through nested JSON parsing. Additional issues include API key leakage (CVE-2026-75760), an infinite loop (CVE-2026-82579), and database schema disclosure (CVE-2026-82580). All vulnerabilities were discovered by researcher PJUllrich using LLM-assisted security research. The vulnerabilities were addressed in version 1.0.0 of ash_ai, released on the same day as the disclosure.

Key Points: • Six critical vulnerabilities disclosed in ash_ai, affecting the Elixir/Ash ecosystem. • CVE-2026-77956 allows remote code execution via user-provided prompts. • All vulnerabilities were patched in version 1.0.0 released on August 30, 2026.