# KPATH > KPATH is the inline enforcement layer for enterprise AI agents. Every governed action passes through its policy enforcement point, which verifies the agent's identity and principal, enforces least-privilege policy, budgets and human approval, can stop one agent or an entire delegation chain, and records everything in a tamper-evident audit log. It governs the request envelope and reads only the payload values a customer declares for a rule. It ships as three products: KPATH Enterprise for the agents an organisation runs on its own systems, KPATH Defence for defence and sovereign environments, and KPATH Developer Edition for the AI coding harnesses a development team uses. KPATH is built by the founders of HYDN Security, is based in the UK, and serves regulated financial services and defence. Observability tells you what happened. KPATH controls what is allowed to happen. KPATH is the policy enforcement point for AI agents. Your identity provider is the policy decision point: it issues agent identities and decides what should be allowed. KPATH acts on that decision inline, at the moment the agent calls, and records what happened. It runs five checks on every call: discover the agents and services that exist, identify the agent and the principal behind it, govern the call against policy, budget and human approval, contain by stopping one agent or a whole delegation chain, and prove with a tamper-evident record an auditor can verify offline. Scope: KPATH governs the request envelope: identity, target, action, size, delegation chain. Of the payload, it reads only the values you declare for a rule, such as a payment amount. Keys: credentials are encrypted under a key held in the customer's own vault or key service. KPATH is a crypto user, not a custodian. Deployment: self-hosted in your own cloud, on premises, air-gapped, or managed by KPATH. Any agent framework. Adoption: Deploy in monitor mode: observe only, enforce no policy, rewrite no agents. Flip to enforce by repointing egress. ## Platform - [What does the KPATH platform do?](https://www.kpath.ai/platform): KPATH governs, audits and controls every interaction your AI agents have with enterprise systems, so every call is identified, checked and recorded. - [How does KPATH give AI agents an identity?](https://www.kpath.ai/platform/identity): How KPATH binds every AI agent call to a verified non-human identity, an owner, a risk tier and a principal, with no credentials held in the agent. - [How does KPATH enforce least-privilege policy on AI agents?](https://www.kpath.ai/platform/policy): How KPATH applies risk-tiered, least-privilege policy to every AI agent call, with inline human approval and your guardrails as a policy step. - [How do you stop a chain of AI agents?](https://www.kpath.ai/platform/chain-kill): Why revoking one AI agent is not enough, and how KPATH stops a whole delegation chain mid-flight and proves the stop held. - [What makes an AI agent audit log evidence?](https://www.kpath.ai/platform/audit-evidence): How KPATH turns an AI agent audit log into evidence: a signed, tamper-evident record, verifiable offline, streamed to your SIEM and held write-once. - [How do you control AI agent spend per agent?](https://www.kpath.ai/platform/budgets): How KPATH enforces AI agent spend limits on the call itself and attributes every unit of spend to an agent and to the originating request. - [How do AI agents discover which services they may call?](https://www.kpath.ai/platform/service-directory): Discovery builds a live inventory from the calls themselves: every agent that calls through KPATH and everything it reaches, with no policy enforced. - [Where does KPATH run?](https://www.kpath.ai/platform/deployment): KPATH runs self-hosted, in a VPC, on-prem, air-gapped or managed, as redundant enforcement points with failure behaviour you choose. ## Learn - [The category, explained plainly.](https://www.kpath.ai/learn): Definitions, sourced analyses of real agent failures, checklists for assessing a vendor, and what the regulations ask of AI agents. - [What is an AI agent control plane?](https://www.kpath.ai/learn/agent-control-plane): The control plane for enterprise AI agents: what it is, the five capabilities it has to deliver, and why a decision is not an enforcement. - [What is AI agent governance?](https://www.kpath.ai/learn/ai-agent-governance): AI agent governance as decision versus enforcement: your identity provider decides, something on the path of the call applies the decision, a record proves it. - [What is an MCP gateway and do you need one?](https://www.kpath.ai/learn/mcp-gateway): What an MCP gateway does, what it leaves uncovered (principal identity, cross-protocol chains, evidence), and where KPATH's governed MCP endpoint fits. - [What is non-human identity for AI agents?](https://www.kpath.ai/learn/non-human-identity): Non-human identity for AI agents: workload identity, principal binding, and why a verified identity is necessary but not sufficient. - [What should an AI agent security platform actually do?](https://www.kpath.ai/learn/agent-security-platform): What an AI agent security platform must do: verify, enforce, stop and prove, plus the ten required controls and the question that separates it from monitoring. - [How does zero trust apply to AI agents?](https://www.kpath.ai/learn/zero-trust-for-ai-agents): How zero trust principles apply to AI agents: verify each call, least privilege per action, assume breach, and where enforcement has to sit. - [How does KPATH map to the OWASP Top 10 for Agentic Applications?](https://www.kpath.ai/learn/owasp-agentic-top-10): The OWASP Top 10 for Agentic Applications (ASI01 to ASI10) mapped to KPATH: where it acts directly, where it limits impact, and what else you need. - [What happened when Replit's AI agent deleted a production database?](https://www.kpath.ai/learn/incidents/replit-production-database-deletion): A sourced analysis of the July 2025 Replit incident: what the agent did, what Replit changed, what remains unknown, and the enforcement lesson. - [What is the Verify-Enforce-Stop-Prove test?](https://www.kpath.ai/learn/verify-enforce-stop-prove): The Verify-Enforce-Stop-Prove test: four framework-neutral questions for any AI agent governance vendor, what pass and fail look like, and how KPATH answers. - [What controls does an enterprise need before AI agents act on real systems?](https://www.kpath.ai/learn/agent-controls-checklist): The required-controls checklist: ten framework-neutral controls an enterprise needs before AI agents act on real systems, with a pass test for each. - [What does the EU AI Act require for AI agents?](https://www.kpath.ai/learn/eu-ai-act-ai-agents): Transparency applies now, high-risk moved to December 2027. What the EU AI Act means for an estate of AI agents. - [What happened in the Amazon Q Developer extension prompt-injection incident?](https://www.kpath.ai/learn/incidents/amazon-q-prompt-injection): A sourced analysis of the July 2025 Amazon Q Developer extension incident: how a wiper prompt shipped, what AWS said, and the enforcement lesson. - [What does DORA mean for AI agents in financial services?](https://www.kpath.ai/learn/dora-ai-agents): What DORA (Regulation (EU) 2022/2554) means for AI agents in financial services: ICT risk management, access control, incident reporting, third-party risk. - [What does the FCA Consumer Duty mean for AI agents?](https://www.kpath.ai/learn/consumer-duty-ai-agents): What the FCA Consumer Duty (PS22/9, PRIN 2A) means for AI agents: Principle 12, cross-cutting rules, four outcomes, monitoring and board reporting. - [How do AI agents change APP fraud exposure and what controls apply?](https://www.kpath.ai/learn/app-fraud-ai-agents): How AI agents change APP fraud exposure under the PSR reimbursement rules from 7 October 2024, and the controls that apply: approval on payments and evidence. - [Did an autonomous agent escape its sandbox at Hugging Face?](https://www.kpath.ai/learn/incidents/hugging-face-sandbox-escape): A sourced analysis of the July 2026 Hugging Face intrusion by an OpenAI evaluation agent: confirmed facts, open questions, and the enforcement lesson. - [What AI agent security incidents should every CISO know about?](https://www.kpath.ai/learn/incidents): A dated table of confirmed AI agent security incidents with primary sources, and how KPATH writes its incident analyses. - [What has gone wrong with MCP security?](https://www.kpath.ai/learn/incidents/mcp-security-2026): A sourced, dated account of what has gone wrong with MCP security since 2025: tool poisoning, critical RCEs, exposed servers, supply chain, and what to enforce. ## Compare - [How should you compare AI agent governance vendors?](https://www.kpath.ai/compare): Four questions to put to any AI agent governance vendor, what a good answer looks like, and when building your own gateway is enough. - [Should you build or buy an AI agent gateway?](https://www.kpath.ai/compare/build-vs-buy): When a gateway plus scripts is enough for AI agent governance, the three conditions under which it stops being enough, and the categories of cost of ownership. ## Trust - [Is it safe to route AI agents through KPATH?](https://www.kpath.ai/trust): What KPATH sees, who holds the keys, where it runs, what happens when it fails, and how its record is verified without trusting KPATH. - [What does KPATH see?](https://www.kpath.ai/trust/what-kpath-sees): The scope KPATH governs, what it never sees, where guardrail inspection runs, and why retention is metadata only. - [How is KPATH deployed and what does each tier entail?](https://www.kpath.ai/trust/deployment-and-architecture): KPATH deployment tiers compared: self-hosted cloud, on premises, air-gapped and managed. Who runs it, where keys live, where evidence lands, network dependency. - [How do you verify a KPATH audit log without trusting KPATH?](https://www.kpath.ai/trust/verify): Checking the KPATH audit record yourself: a signed, tamper-evident export, a standalone verifier, a sample export to try first, and what the result proves. ## Developer - [How do you govern a team's AI coding agents?](https://www.kpath.ai/developer): KPATH Developer Edition governs Claude Code, Cursor and GitHub Copilot, and what they reach: identity per developer, policy, one governed MCP endpoint. - [How do you control what Claude Code can access?](https://www.kpath.ai/developer/claude-code): What a team can control about Claude Code: per-developer policy on what the harness reaches, credentials it never holds, and a record off the laptop. - [How do you control what Cursor's agent can access?](https://www.kpath.ai/developer/cursor): What a team can control about Cursor's agent: per-developer policy on what it reaches, credentials the agent never holds, and a verifiable record. - [How do you control what GitHub Copilot's agent can access?](https://www.kpath.ai/developer/copilot): What a team can control about GitHub Copilot's agent: per-developer policy on what it reaches, credentials it never holds, and a verifiable record. - [What is a governed MCP endpoint?](https://www.kpath.ai/developer/mcp-endpoint): What a governed MCP endpoint is: one route, fan-out to approved tools, per-developer policy and a tamper-evident record, and how it differs from an MCP gateway. ## Solutions - [Where agents act on things that matter.](https://www.kpath.ai/solutions): KPATH in regulated financial services, in platform engineering, and as a neutral enforcement layer for sovereign environments. - [How do banks run AI agents safely?](https://www.kpath.ai/solutions/financial-services): Banks and insurers are putting agents on AML, sanctions, KYC, card fraud and claims. KPATH sits between those agents and your core systems. - [How does a platform team govern every AI agent and coding harness in the estate?](https://www.kpath.ai/solutions/platform-engineering): How a platform team uses KPATH Developer Edition as the beachhead, a monitor-mode inventory to find every agent, and one governed MCP endpoint per team. - [What is KPATH Defence?](https://www.kpath.ai/solutions/defence): KPATH Defence: a neutral enforcement and accountability layer for defence and sovereign environments. Human control, air-gapped, keys in your custody. - [What is agentic commerce and how do you give customer agents a governed way in?](https://www.kpath.ai/solutions/agentic-commerce): Customer and partner agents are already arriving. Give them a governed way in: identity, policy, spending limits and a record of every action. ## Consultancy - [What does the KPATH consultancy practice do?](https://www.kpath.ai/consulting): KPATH consultancy: moving AI from pilots to production across governance, security and operations, from the team that builds the control plane. ## Services - [What the consultancy practice covers.](https://www.kpath.ai/services): Strategic advisory, AI security and governance, agentic frameworks, and automation and AI readiness, from the team that built HYDN Security. - [What does KPATH's AI strategic advisory cover?](https://www.kpath.ai/services/strategic-advisory): AI strategic advisory for leadership teams: operating model design, build vs buy vs partner, where agents create advantage, and avoiding lock-in. - [What does KPATH's AI security and governance advisory cover?](https://www.kpath.ai/services/ai-security-governance): Agent threat modelling, prompt injection and tool misuse, least privilege for non-human callers, and the evidence design your risk function will accept. - [How does KPATH help get an agent from demo to production?](https://www.kpath.ai/services/agentic-frameworks): Agent architecture, role and tool surface design, framework selection, evaluation and cost control for teams taking agents from prototype to production. - [How do we know whether our processes are ready to automate?](https://www.kpath.ai/services/automation-ai-readiness): An honest readiness assessment across data, process, systems and people, with a scored shortlist of automation candidates and a sequenced roadmap. ## Company - [Who builds KPATH?](https://www.kpath.ai/company): The founders, advisor and UK research partners behind KPATH, and the HYDN Security record the team brings to AI agents. - [Why does KPATH exist?](https://www.kpath.ai/company/why-kpath): The gap KPATH was built for: decisions without enforcement at the moment of action. The founders' background, the two lanes, and what neutral means here. ## Insights - [Notes from the people building it.](https://www.kpath.ai/blog): Analysis from the KPATH team on agent containment, MCP security, what a control plane actually does, and what the EU AI Act changed. - [AI agent kill switch: why stopping one agent isn't enough](https://www.kpath.ai/blog/ai-agent-kill-switch-why-stopping-one-agent-isn-t-enough): Why an AI agent kill switch that stops one agent leaves the agents it started running, and the five things whole-chain containment needs. - [The EU AI Act just passed a milestone. Here is what actually changed](https://www.kpath.ai/blog/the-eu-ai-act-just-passed-a-milestone-here-is-what-actually-changed): What came into force in the EU AI Act on 2 August 2026, what the Digital Omnibus deferred, and the inventory question every path through it runs into. - [MCP security: what has gone wrong in 2026](https://www.kpath.ai/blog/mcp-security-what-has-gone-wrong-in-2026): What broke in the Model Context Protocol ecosystem in 2026, why agents fail differently from ordinary software, and six things to do about it. - [What an AI agent control plane actually does](https://www.kpath.ai/blog/what-an-ai-agent-control-plane-actually-does): The five things that have to happen between an agent's request and its action: intent arbitration, route selection, per-action policy, context continuity. ## Machine-readable - [Full text of every page](https://www.kpath.ai/llms-full.txt): the complete content of the site as one file. - [Sitemap](https://www.kpath.ai/sitemap-index.xml)