← back

CVE-2026-59726

CRITICAL · 10.0 EPSS 0.5%
hype LIKELY HACK · 72 hack

Patch released, real vuln, mass media coverage, no confirmed exploitation; likely HACK.

What: Ruflo (Claude Code/Codex agent harness) versions <3.16.3 expose unauthenticated MCP bridge endpoints allowing remote code execution, shell access, API key theft, and agent memory poisoning. CVSS 10.0.

Why it matters: Patch released 21 days ago (3.16.3 fixes it); mass coverage across security media and developer platforms; no KEV listing yet but threat is real — unauthenticated RCE on a system touching AI credentials and agent state. Active chatter shows defenders and teams aware of exposure window.

Where it's seen: Bluesky posts from security researchers and outlets (HackerNews, OffSeq, BlindThoughts); links to threat intel; multilingual discussion; upgrade guidance circulating. No confirmed in-the-wild exploitation reported, but CVSS 10.0 and simple attack surface drive urgency.

RISK: CRITICAL — Unauthenticated RCE exposing API keys and agent memory; trivial network attack surface.

Generated by claude-haiku-4-5 from public posts and authoritative metadata. AI can make mistakes — verify against vendor advisories before acting. 7/30/2026, 11:53:08 AM

Description

Ruflo is an agent meta-harness for Claude Code and Codex. Prior to 3.16.3, ruflo's default docker-compose deployment exposed the MCP bridge POST /mcp and POST /mcp/:group endpoints without authentication, allowing an unauthenticated network attacker to invoke tools/call to terminal_execute, obtain a shell in the bridge container, read provider API keys, and poison AgentDB learning-store patterns. This issue is fixed in version 3.16.3.

CVSS 3.1 breakdown

Exploitability 3.9 · Impact 6.0
vector CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:C/C:H/I:H/A:H
Attack vector
Network
Complexity
Low
Privileges required
None
User interaction
None
Scope
Changed
Confidentiality
High
Integrity
High
Availability
High