June 2026 Summaries

MCP gateways are becoming crucial in enterprise AI deployments, yet they often lack necessary security controls, leading to significant vulnerabilities. The primary risks include inadequate tool permissions, insufficient user scoping, unannounced vendor-side changes, fragmented policies across AI clients, and exposure of sensitive data. MCP gateways serve as policy enforcement layers, controlling access and protecting data between AI agents and MCP servers. Effective MCP security requires per-tool policy enforcement, identity-driven access control, change management workflows, and real-time data protection, with policies ideally being code-based for better management and auditability. Security teams must ensure these controls are in place to mitigate the inherent risks associated with MCP deployments, emphasizing the need for a centralized control plane and comprehensive policy lifecycle management.

Mary, a senior engineer at a mid-size SaaS company, faced a significant unexpected expense when her team inadvertently incurred a $15,000 monthly bill due to a configuration change that routed their SRE agent's operations through a costly API without budget controls. This situation highlights a broader issue where many organizations lack comprehensive visibility over their AI-related costs, leading to surprise invoices when token and cloud computing budgets are exceeded. Barndoor, a tool designed to manage and optimize AI expenditures, could have mitigated this by implementing budget controls, rate limits, and appropriate model access policies. By setting budget ceilings, enforcing rate limits, and choosing cost-effective models, Barndoor allows organizations to manage AI costs proactively, preventing overspending and surprise bills.

Prompt caching offers significant cost savings in AI products, providing up to a 10x discount on the largest expenses associated with AI usage by reusing previously processed text at a reduced price. However, the system can fail silently, leading to unexpected cost increases without any direct error notifications, as highlighted by a case where a client's API key incurred four times the usual cost due to caching issues. These failures are often caused by discrepancies in caching implementation across different vendors, as well as data protection measures that inadvertently alter requests, negating caching benefits. Barndoor LLM Gateway addresses these challenges by automating cache instructions in the correct format for each vendor, ensuring data protection does not interfere with caching, and monitoring every request to detect issues promptly. This solution allows companies to manage AI costs proactively, without requiring engineers to become caching experts or compromising on data security, thereby preventing unexpected financial surprises at the end of the month.

Employees are independently adopting AI tools to enhance productivity, often bypassing IT department permissions, leading to a fragmented work environment where AI-driven tasks are integrated into multiple systems such as email, messaging, and databases. This widespread and unsanctioned use of AI tools creates significant challenges for IT departments, particularly in tracking and managing these tools as they operate outside corporate oversight. The traditional Identity and Access Management (IAM) systems are inadequate for monitoring AI agents, which can perform tasks without human judgment, potentially causing data mishaps. Companies face the dilemma of balancing the benefits of AI with the need for governance and security, as AI agents can autonomously execute actions that could have serious implications if not properly controlled. Solutions like Barndoor are emerging to offer a centralized control plane, allowing organizations to govern AI tool interactions and maintain data security within established boundaries.

Barndoor has enhanced its agent governance capabilities, extending them across Multi-Cloud Platforms (MCPs) and Large Language Models (LLMs). The update introduces new controls, such as tighter access permissions for tools and the addition of eight new MCPs, including Google and Sardine. The Barndoor LLM Gateway now centralizes management of LLMs and enforces data protection policies that extend to MCPs, enabling centralized setting of access, routing, and usage policies, along with API key management and data redaction. The updates also include improvements in agent lifecycle management, tool permission categorization using CRUD visibility, and enhanced human-in-the-loop workflows for write operations, allowing admins more precise control over agent actions. Additionally, Barndoor's data protection features now apply uniformly across MCPs and LLMs to safeguard sensitive data and prevent unauthorized access, while new features for policy setup and agent registration enhance operational security and efficiency.