The category, explained plainly.
Definitions, sourced analyses of real agent failures, checklists for assessing a vendor, and what the regulations ask of AI agents.
Learn
What is an AI agent control plane?
Agents are scaling into production faster than the controls to govern them. A control plane closes that gap: the layer that sees every agent, decides what each one is allowed to do, and enforces that decision on live traffic. KPATH is that control plane.
What is AI agent governance?
AI agent governance is the set of decisions about what each agent may do, on whose authority, and within what limits, together with the means to apply those decisions when an agent acts and to prove afterwards that they were applied. The decision and the enforcement are separate jobs, and most programmes only have the first.
What is an MCP gateway and do you need one?
An MCP gateway is a single endpoint in front of your Model Context Protocol servers, so clients connect to one place, tools are inventoried, and tokens are checked at the boundary. If you run more than a handful of MCP servers you need something there. It does not govern the principal, the chain across protocols, or the evidence.
What is non-human identity for AI agents?
Non-human identity is the practice of giving software callers, including AI agents, a verifiable identity of their own rather than a shared key or a borrowed human login. For agents it goes further: the identity must be bound to the person or system the agent acts for, and something must enforce what that identity may do on each call.
How does zero trust apply to AI agents?
The principles hold. Nothing is trusted because of where it sits, every request is checked against policy, access is the least the job needs, and you plan for compromise. What changes is the subject. An agent is a caller of its own, it acts for someone else, and it hands work to other agents, so the check has to happen on every call it makes, not once at login.
What should an AI agent security platform actually do?
It should verify who is calling and for whom, enforce policy on the call before the action happens, stop an agent or a whole chain and keep it stopped, and produce a record a third party can verify. Anything that only watches and reports is monitoring with a new name. The required-controls checklist names the ten controls.
How does KPATH map to the OWASP Top 10 for Agentic Applications?
KPATH acts directly on five of the ten risks: tool misuse, identity and privilege abuse, insecure inter-agent communication, cascading failures and rogue agents. On four more it limits what a successful attack can reach. Code execution inside the agent is for your sandbox. Each risk below says which, and what else you need.
What is the Verify-Enforce-Stop-Prove test?
The Verify-Enforce-Stop-Prove test is four questions for any AI agent governance vendor: can it verify who is calling and for whom, enforce the decision before the action, stop a whole chain and keep it stopped, and prove what happened to someone who does not trust its console. Each has a pass and a fail you can check.
What controls does an enterprise need before AI agents act on real systems?
Ten controls, each testable before go-live: identity bound to principal, least privilege per action, budget on the call, human approval inline, whole-chain stop, tamper-evident record, monitor before enforce, key custody kept by you, envelope-only scope, and failure behaviour decided in advance. Together they are the required-controls checklist.
What does the EU AI Act require for AI agents?
The Digital Omnibus pushed the high-risk obligations back to December 2027, and a lot of coverage read that as the Act being delayed. It was not. Transparency duties, enforcement powers and the penalty framework all arrived on 2 August 2026.
What does DORA mean for AI agents in financial services?
DORA, Regulation (EU) 2022/2554, has applied to EU financial entities since 17 January 2025. It does not mention AI agents. It requires ICT risk management owned by the management body, access limited to approved functions with strong authentication, every ICT incident recorded and major ones reported, and full responsibility for third-party ICT services. Agents are inside all four.
What does the FCA Consumer Duty mean for AI agents?
The Consumer Duty requires a firm to act to deliver good outcomes for retail customers, avoid foreseeable harm, and monitor and evidence outcomes, with an annual report to the governing body. An AI agent that serves, advises or acts for a retail customer is inside the Duty, and the firm must show what it did and could stop it.
How do AI agents change APP fraud exposure and what controls apply?
Since 7 October 2024, UK payment firms must reimburse victims of authorised push payment fraud on Faster Payments and CHAPS up to £85,000 per claim, within five business days, split 50:50 between sending and receiving firm. An agent that moves money can be manipulated into an authorised payment at machine speed. The controls: human approval on payments, and evidence.
Incidents
Updated
Start in monitor mode. See everything before you enforce anything.
Deploy in monitor mode: observe only, enforce no policy, rewrite no agents. Flip to enforce by repointing egress. See the agents already running in your estate before you decide what to enforce.