Hello!

Welcome to Secure Prompt’s weekly newsletter, issue #13.

This week confirms a clear shift: attackers are no longer just targeting AI models — they’re targeting agent infrastructure, orchestration layers, and developer tooling.

From hijacked Ollama and vLLM endpoints being resold as a service, to malicious Moltbot/Clawdbot lookalikes and unauthenticated MCP servers enabling remote access, AI systems are becoming the new exposed perimeter.

The takeaway is blunt:
If an AI system can act, persist memory, or invoke tools - it must be secured like production infrastructure, not a demo.