CVE-2026-48710 Starlette Host-Header Auth Bypass
Summary
A critical host-header authentication bypass vulnerability (CVE-2026-48710) in Starlette and FastAPI affects many Python ASGI applications, including AI inference servers (e.g., vLLM), AI proxy servers (e.g., LiteLLM), and MCP gateways, potentially allowing unauthorized access.
View Cached Full Text
Cached at: 05/27/26, 09:30 AM
Similar Articles
Millions of AI agents imperiled by critical vulnerability in open source package
A critical vulnerability (CVE-2026-48710, named BadHost) in the open-source ASGI framework Starlette exposes millions of AI agents and servers to potential data theft and credential compromise, affecting frameworks like FastAPI, vLLM, and LiteLLM. Patched in Starlette 1.0.1, the flaw is trivial to exploit and underscores risks in the AI tooling ecosystem.
CVE-2026-48710: A Maintainer's Perspective
Marcelo Trylesinski shares his perspective on CVE-2026-48710, a security vulnerability in Starlette involving path-based authorization bypass via manipulated Host headers. He argues the vulnerability stems from application patterns and deployment, not the framework itself.
Anthropic Claude Code Leak Reveals Critical Command Injection Vulnerabilities
Critical command injection vulnerabilities (CVE-2026-35022, CVSS 9.8) discovered in Anthropic's Claude Code CLI and SDK allow attackers to execute arbitrary commands and steal credentials through environment variables, file paths, and authentication helpers. The flaws enable poisoned pipeline execution attacks in CI/CD environments, requiring immediate patching and configuration changes.
@Star_Knight12: Next.js just got its worst vulnerability ever, CVSS 8.6. → affects versions 13.4.13+, 14.x, 15.x, and 16.0.0–16.2.4 → a…
Next.js has a critical vulnerability (CVSS 8.6) affecting versions 13.4.13+, 14.x, 15.x, and 16.0.0–16.2.4, allowing unauthenticated attackers to access internal services, cloud credentials, and API keys. Upgrade to 15.5.16 or 16.2.5 immediately.
Two Claude Code sandbox bypasses in five months, both fixed silently. What does the shared responsibility model look like for AI agents?
Researcher Aonan Guan disclosed a second Claude Code network sandbox bypass via HackerOne, exploiting a SOCKS5 hostname null-byte injection. The vulnerability affected versions 2.0.24 through 2.1.89 and was fixed silently twice, raising questions about the shared responsibility model for AI agents.