Niels Provos Deploys IronCurtain To Block Rogue Agents
Veteran security engineer Niels Provos is developing IronCurtain, an open-source tool to stop autonomous LLM agents from taking unauthorized actions. It routes agent-generated TypeScript and MCP tool-call requests through a trusted proxy policy engine that enforces a user-written 'constitution', verifies rules with a compiler LLM and test scenarios, and blocks or escalates requests. The early research release invites developer testing and feedback.
Key Points
- 1Routes agent tool calls through a trusted MCP proxy that enforces policy and human escalation
- 2Uses a user-written 'constitution' compiled by an LLM with verification and test-case generation
- 3Prevents agents accessing files, credentials, or policy files, enabling safer autonomous workflows and audits
Scoring Rationale
New open-source enforcement approach provides actionable protection; early-stage research and limited evaluation constrain near-term enterprise adoption.
Sources
Public references used for this report.
Practice interview problems based on real data
1,625 SQL & Python problems across 15 industry datasets — the exact type of data you work with.
Try 250 free problems
