NewCore ads

Ask three people in your company how many AI agents are running and you'll get three different answers. That's the wrong question anyway. Agents don’t arrive like employees or software. A sales team turns on the agent built into their CRM. An engineer wires one into a workflow. Someone gives their AI assistant access to company email or calendar. None of it goes through IT or HR. The question that matters for each agent: who does it act for, where does it run, what can it touch, and who’s accountable for it? → A personal agent should never reach more than its owner can. → A shared agent, like the HR agent anyone can ask, may hold broad access but should give each user only what they're entitled to. → An autonomous agent should get only what its task needs. None of this is possible for an agent you haven't found. It starts with discovery: finding every agent, including the ones no one approved, tracing each one to the person accountable for it, and seeing what it can reach. Here's how we do it at NewCore: https://lnkd.in/dqvRBTkk #AIAgents #IdentitySecurity #AgenticIdentity

Cybersecurity leaders must establish authoritative identities for AI agents as software workloads in order to safely scale AI automation and securely enable agentic initiatives. Identity-based controls are the foundational control plane for organizations seeking to securely enable agentic AI. Get the full report.

Identity-based controls are the foundational control plane for organizations seeking to securely enable agentic AI. Get the full report.

MCP solved how agents talk to your systems. It did not solve whether they should. That is not a criticism of the protocol. It was built to standardize the connection, and it did. Every tool now looks the same to every agent, which is exactly why adoption moved so fast this year. But a standard connection is an entry ticket. It gets the agent to the door of your CRM, your data warehouse, your ticketing system. It says nothing about whether this agent, acting for this person, should be allowed through that door for this action, at this moment. Most stacks I see answer that with a gateway that logs. Every call recorded, nothing decided. That is a camera, not a lock. And one prompt injection turns a standing credential into access to everything it could ever reach, while the log faithfully records the whole thing. The decision has to sit in the path. Evaluate the policy per call. If the call is allowed, mint a credential that lives as long as the task and no longer, tied to a human who can be named. If it is not allowed, the call does not happen. Connect contextually, not perpetually. The protocol gets you connected. Enforcement is the point. Read more on how we approach it: https://lnkd.in/duXrum7e NewCore #MCP #Security

Ask your security team one question this week: When an AI agent in your environment does something wrong, who answers for it? I’ve asked this in a lot of rooms lately. The answer is almost always “we have the logs.” Logs tell you what happened. They don’t tell you who allowed it, under which policy, or whether anyone could have stopped it. “The AI did it” is not an answer a CISO can give a regulator, a board, or a customer. Someone owns every action. If you can’t name that someone BEFORE the agent runs, you’ll be reconstructing it afterwards, under pressure, from logs that were never built to answer the question. Accountability has to be designed in: A named owner for every agent. A policy that decides what it can do, enforced the moment it acts. A real-time escalation to that owner when the engine can’t decide. Stop watching agents. Start governing them. Gartner on securing AI agents with identity governance: https://lnkd.in/dekeyNxh

Cybersecurity leaders must establish authoritative identities for AI agents as software workloads in order to safely scale AI automation and securely enable agentic initiatives. Get the full report.