Fetching from the wire…
Security2026-08-29 · source-backed
CVE-2026-55830 breaks the mechanism the whole library rests on. Sandboxed code is rewritten so attribute access goes through _getattr_, item access through _getitem_, writes through _write_. Argument-name validation rejects those protected names for regular args, *args, **kwargs and keyword-only args, and misses positional-only. So def f(_getattr_=evil, /) makes the hook a local, and every rewritten access calls the attacker's function. Shadowing _print_ additionally captures the internal _getattr_ hook RestrictedPython passes in. Full policy escape from a syntax feature the validator forgot existed. (GitHub Advisory)
Each link below shares sources, entities, or timing with this story.
CVE-2026-45018 covers Chainlit >=2.4.0rc0 <2.12.0. With features.mcp.enabled = true, POST /mcp accepts a user-controlled fullCommand for stdio transport. validate_mcp_command() checks the executable name against an allowlist and never inspects arguments, so npx -y -c '<command...
CVE-2026-55641: the request guard decides a caller is local by reading the client-controlled Host header, and the server binds 0.0.0.0 by default while the CLI prints "localhost." Any remote attacker sending Host: localhost gets /v1 proxy access with no API key, no CLI token a...
NVD posted nine advisories on August 25, clustering into one shape: a local server assuming a browser can't reach it. PraisonAI validated MCP origins with request_origin.startswith(allowed) against a localhost allowlist, so an attacker-registered localhost.attacker.com passes...
— Security researcher Ari Marzouk disclosed 30+ vulnerabilities (24 CVEs) affecting Cursor, GitHub Copilot, Windsurf, Zed, Kiro, Roo Code, Junie, and Cline. The devastating finding: every tested AI IDE is vulnerable because none accounts for autonomous LLM agent behavior in th...
CVE-2026-55546 at 9.8 sits in verify_math_expression() in QWED-MCP, described by its authors as "a deterministic verification gateway for MCP." It hands the attacker-controlled expression and claimed_result to parse_expr() after normalizing caret syntax, with no global_dict re...
A novel vulnerability in OpenClaw: the current working directory path was embedded into agent system prompts without sanitization. Control characters in directory names (newlines, Unicode bidi/zero-width markers) break prompt structure and inject attacker instructions. Patched...
MindPattern daily
One email a day at 7 AM. Sources and a take on every story. Unsubscribe anytime.