1 CRIT · 9 HIGH · 2 MED · THREAT RED · 12 items · Generated in 232s
Unauthenticated attackers can read, write, and delete stored memory data remotely, breaking confidentiality and integrity.
Patched MCP Atlassian can still be driven to reach cloud metadata or internal services via DNS rebinding SSRF.
Unauthenticated requests can trigger shell command execution as the server process, leading to full host compromise in exposed setups.
Exposed Gradio apps can be abused to reach internal services and steal cloud credentials via SSRF.
Remote clients can bypass token checks and access the default Telegram session without a valid bearer token.
A hook token could trigger CLI runs with owner-only MCP tool access, breaking isolation and enabling privileged actions.
An authenticated MCP client or manipulated agent can read sensitive local files and exfiltrate them as Confluence attachments.
Auth bypass let unauthenticated requests reach MCP tools, exposing connected capabilities until LiteLLM 1.84.0.
Hidden prompt injection can make coding agents expose repo secrets and commit them into code.
AI coding agents can trick approvals, letting risky actions pass human checks and weakening basic Unix trust boundaries.
Unauthorized access exposed a full airline booking database, risking customer data theft and large-scale privacy impact.
Shows computer-use agents can be tricked by UI race conditions into taking unintended actions.