November 2026 Summaries
5 posts from Barndoor
Filter
Month:
Year:
Post Summaries
Back to Blog
Barndoor has made its AI Gateway and Automations platform available through Google Cloud Marketplace, enabling enterprises to purchase it using existing Google Cloud billing. The platform is designed to govern agentic AI by controlling access to MCP servers and tools, inspecting and protecting data in tool inputs and outputs, and maintaining audit trails of actions. Its LLM Gateway centralizes model routing, virtual key management, budget controls, and prompt and response inspection to help prevent sensitive-data leaks and manage AI spending. Barndoor Automations supports repeatable workflows that use AI selectively while applying existing access controls and allowing authorized users to invoke workflows from tools such as Slack or Claude. Google Cloud said the marketplace availability is intended to help customers deploy and scale Barndoor’s governance capabilities on its infrastructure.
Nov 12, 2026
549 words in the original blog post.
Major security platforms including CrowdStrike, Palo Alto Networks, Cisco, SentinelOne, and Check Point are expanding into AI security through acquisitions that add capabilities such as model testing, attack detection, prompt protection, monitoring, and agent observability. The material distinguishes AI security, which focuses on threats such as prompt injection, jailbreaks, and data theft, from AI governance, which determines and records who may use particular models, tools, data, actions, and budgets. It frames governance as four functions—testing before deployment, production monitoring, runtime enforcement, and audit documentation—and argues that security suites commonly perform the first two well but often lack inline controls that evaluate every model request, MCP tool call, or agent action before it occurs. Key gaps include separate permissions for agents, tool- and action-level controls, DLP for tool-call data, hard spending limits, coverage of unmanaged AI connections, and logs that link actions to governing policies. The text presents Barndoor as an AI gateway designed to fill these enforcement gaps through per-request policy checks, delegated agent identities, data protection, spend controls, and audit logs that can feed existing SIEM and GRC systems.
Nov 10, 2026
2,778 words in the original blog post.
Barndoor Automations and its AI Gateway are presented as a governance-focused alternative to ungoverned AI agents, emphasizing constrained access, schema-validated outputs, named identities, policy enforcement, and auditable execution records around model calls. In a test task that classified customer support messages as bugs, billing issues, or feature requests, required input validation blocked blank messages, output schemas restricted labels to approved values, and gateway policies denied calls by default until explicitly allowed, with audit logs recording users, agents, statuses, and trace IDs. Workflows are defined in reviewable FML Blueprints that can be deployed in Barndoor’s cloud or through the open-source Frags runtime in a customer-controlled environment, and can be accessed through APIs or MCP. The evaluation also found a limitation: while structural validation handled empty or malformed inputs, a meaningless but valid string such as a single period still reached the model and received a classification, indicating that semantic input checks require gateway screening or workflow-specific instructions.
Nov 05, 2026
860 words in the original blog post.
Barndoor has launched a two-agent system for creating governed AI automations, separating planning from execution to reduce misunderstandings and unpredictable results. Its design agent converts natural-language requests into a requirements brief and a reviewable Blueprint sketch that identifies answered, assumed, unknown, or irrelevant details, while its build agent turns approved plans into FML workflows and validates them against the design. The platform aims to address limitations of conventional and newer AI automation tools, including rigid logic, coding requirements, unreliable judgment calls, limited auditability, and weak identity, data, spending, and access controls. Built on the open-source Frags workflow engine and its Frags Modeling Language, Barndoor Automations restricts AI involvement to authorized tasks and applies existing enterprise governance policies throughout each workflow. The company says its governed content-marketing Blueprint can save teams 20 to 30 minutes weekly and offers enterprises a free trial to build production-ready automations.
Nov 04, 2026
793 words in the original blog post.
Claude Code v2.1.316, released on October 31, 2026, strengthened its apps gateway by introducing layered policy evaluation, shared group budgets, Amazon Bedrock guardrail input tagging, and validation of every model named in a request. The discussion argues that layered policies reduce the risks of first-match rules, where broad permissions can override narrower restrictions because of rule order, while individual and pooled spending limits can limit damage from leaked credentials or runaway agents. It also emphasizes that guardrails and data-loss protections must be correctly integrated into the request path, including required prompt tags, and that gateways should inspect all requested models to prevent allow-list bypasses. While recognizing these changes as useful for Claude Code traffic, the text contends that organizations using multiple AI providers, applications, agents, and MCP servers need centralized policies, budgets, credential management, auditing, and enforcement across all model and tool interactions. It presents Barndoor as an enterprise platform intended to provide these unified LLM and MCP governance capabilities, including identity-based access controls, agent-specific profiles, inline data protection, virtual keys, spend controls, and SIEM audit logging.
Nov 02, 2026
2,098 words in the original blog post.