BREAKING
Attackers Scan Web for MCP Servers
0
IPs
distinct sources
0
days
log window
0
AI-related requests
Recon Ahead of Adoption
What Attackers Probed
1
POST /mcp handshakes
↓
2
.claude & .cursor secrets
↓
3
Exposed Ollama scans
Weak Spots vs Defenses
Weak Points
●
Plaintext long-lived tokens
●
Optional mutual auth
●
Low security visibility
Defenses
●
Mutual TLS
●
Short-lived tokens
●
Log and monitor endpoints
Assume Exposure From Day One
AI NEWS BLITZ
SANS finds adversaries are already probing the web for AI assistant infrastructure.