1 CRIT · 7 HIGH · 2 MED · 2 INFO · THREAT RED · 12 items · Generated in 228s
Patch bypass restores unauthenticated remote code execution by letting npx auto-install and run attacker-chosen packages.
Unsanitized MCP input can inject Stata shell commands and achieve OS-level code execution on the server account.
Scoped reads can leak another tenant's data when namespace prefixes overlap in affected LangGraph stores.
An authenticated user can use server-side fetches to access internal services or cloud metadata and exfiltrate sensitive data.
Unauthenticated regex input can trigger CPU exhaustion and stall structured-output requests in vLLM until upgraded.
A crafted username can run local shell commands as the MCP server user, enabling host compromise or data theft.
Unauthenticated path traversal can expose session log files and leak sensitive data from the server filesystem.
Unauthenticated requests can leak system and environment details that help attackers map and target the AI service.
NIST exploring AI for rising vuln volume signals process shifts, but this item reports no direct threat or exploit.
Autonomous agents targeting critical infrastructure could speed up intrusions and raise the risk of real-world disruption.
A poisoned dependency could steal secrets from build and runtime hosts, enabling broader compromise across many organizations.
Vendor announcement about AI security readiness features, with no specific vulnerability, exploit, or incident described.