1 CRIT · 7 HIGH · 1 MED · 3 INFO · THREAT RED · 12 items · Generated in 219s
Unauthenticated attackers can read, add, or delete stored memories through exposed document API routes.
Lets an attacker bypass the SSRF fix and reach cloud metadata or internal services via DNS rebinding.
Unauthenticated requests can run OS commands as the server process, enabling full host compromise where the MCP endpoint is reachable.
SSRF can expose internal services and cloud credentials, and open redirects can aid phishing or chained attacks.
Unauthenticated clients can hijack the default Telegram session via path traversal and access that account's MCP tools.
A valid hook token could access owner-only MCP tools, breaking isolation and enabling privileged actions.
An attacker or manipulated agent can read local files the MCP server can access and exfiltrate them to Confluence.
Auth bypass let unauthenticated requests reach MCP tooling, exposing downstream tools and data until 1.84.0.
Third-party agent skills can hide malicious logic or chained flaws that compromise AI agents and connected systems.
This is a policy and trust warning about possible data forwarding, not a confirmed technical vulnerability with public evidence.
This is industry news about AI-assisted bug finding and supply chain defense, not a disclosed vulnerability or active exploit.
General security marketing about reducing AI risk exposure; no specific incident, exploit, or vulnerability is described.