PRIMARY-SOURCE REVIEW · 2026-11-18T09:00:00+09:00

Arize Phoenix: Format string injection in applyChatTemplate leaking server secrets.

In AI model evaluation and prompt tracking framework Arize Phoenix, a format string injection vulnerability in the applyChatTemplate GraphQL endpoint (CVE-2026-53998 / GHSA-gfv3-m769-5xm3) was fixed. Evaluating user-supplied prompt templates processed Python format strings without validation, risking leakage of server environment variables, encryption keys, and internal object attributes. Fixed in version 15.7.0.

What to check

Upgrade arize-phoenix to 15.7.0 or later. Confirm access controls and authentication settings on externally exposed Phoenix UI and GraphQL endpoints.

A first fixed version is listed only when the advisory provides one. Advisory counts are not incident counts.

Primary sources

GHSA-gfv3-m769-5xm3 / Arize Phoenix 15.7.0

Related topics

libp2p: Fixes for gossipsub topic map exhaustion DoS and ASN.1 DER decoder stall vulnerabilities. · SimpleSAMLphp SAML2: Authentication bypass via HTTP-Artifact signature validation omission (incomplete fix of CVE-2026-49283). · EGroupware: Authentication redirection vulnerability via user-controlled external IdP parameter in SAML authentication. · Next.js 16.4.5: Cross-parameter cache leak in nested use cache functions via delayed inner stream metadata. · AWS Advanced JDBC Wrapper: Java deserialization RCE via cache poisoning. · Filament Curator: Stored XSS via inline SVG serving and multi-tenant picker search disclosure. · qbee-agent 2026.27: Local privilege escalation via symlink following in createFile helper. · CIRCUTOR SCADA WAVE and Rucio: Missing JWT signature validation and token verification bypass (CVSS 10.0). · 126 malicious npm packages: Typosquatting and developer credential harvesting.