Varonis Expands Atlas AI Gateway With LLM and MCP Gateways
Summary
Varonis says its Atlas AI Gateway now includes LLM and MCP gateways alongside runtime guardrails, offering centralized model and token controls, tool authorization, identity validation, prompt and response inspection, and activity logging.
Key points
- Varonis says Atlas AI Gateway now combines runtime guardrails with built-in LLM and MCP Gateways.
- The LLM Gateway is described as centralizing model access, usage visibility, and token-spend controls across more than 95 providers and 3,000 models.
- Organizations can set model and usage policies, budgets, limits, and alerts by user, group, model, provider, or endpoint.
- The MCP Gateway governs which tools agents and MCP servers can use and what actions they may perform, including read-only versus write access.
- Varonis says the MCP Gateway validates identity through an IdP, supports delegated credentials, and records tool calls alongside prompts for investigation.
- Runtime guardrails inspect prompts and responses and can alert, modify, or block them under policy; the article says they are intended to address risks including sensitive-data exposure and prompt injection.
- The article recommends inventorying AI systems and sensitive-data access, governing models and tools, applying limits and guardrails, and monitoring activity through dashboards and SIEM/SOAR integrations.
Article Details
- Event Type
- Product capability expansion: Varonis expanded Atlas AI Gateway with LLM and MCP Gateways.
- Impact
- The gateways are presented as a centralized way to govern AI model access, token usage and spend, prompts and responses, and agent and MCP tool calls. The article describes intended security and cost-control capabilities; it reports no specific incident or realized impact.
Vendors
Products
Atlas AI GatewayWhat Atlas AI Gateway deliversLLM GatewayLLM Gateway: Governs model access and token spend, which otherwise would sprawl across providers with no visibility into usage or cost.MCP GatewayMCP Gateway: Governs the tools agents use to act, which would otherwise be called without central control.