Levo — Launch Week 2026 adds AI firewall + MCP security testing
• Category: Security
AI relevance: These releases target the AI runtime layer (prompts, retrieval, tool calls, and MCP servers), directly addressing the security controls needed to operate agents in production.
- AI Firewall is positioned as an inline control plane that inspects prompts, retrieval, and tool calls to block jailbreaks, prompt injection, model extraction, and data exfiltration.
- AI Gateway focuses on governance of third-party AI tools (identity binding, prompt inspection, allowlists/quotas, and centralized credentials).
- MCP Discovery inventories MCP servers across laptops, clouds, and remote environments, then scores risk based on exposure and data/privilege access.
- MCP Security Testing runs agent-style conversations to detect token mismanagement, privilege escalation, command injection, and prompt injection in MCP toolchains.
- Levo also adds integrations for LiteLLM and Portkey to insert runtime guardrails without redesigning the AI stack.
Why it matters
- Enterprises are rolling out agents faster than they can inventory or test MCP servers; discovery + testing closes a growing visibility gap.
- Inline guardrails for prompts and tool calls are the practical controls needed to move agent workloads from pilot to production.
What to do
- Map your AI runtime surface: list every MCP server, agent, and proxy in use, then assign owners.
- Insert a policy choke point: route prompts/tool calls through an enforceable policy layer (firewall or gateway).
- Test MCP tools like apps: run adversarial tests that mimic indirect prompt-injection paths.