- George Kurtz of CrowdStrike said on No Priors that agents are given a user's identity and free rein to download packages, with little oversight.
- He called agents a black box. Our reading is that they need the same identity limits and review as any new worker.
George Kurtz, CEO of CrowdStrike, argued on the No Priors podcast that companies trust AI agents too freely. In his account, an agent gets a person's identity and permission to download packages. Then nobody watches the result. This is his view, offered as a warning.
What was said
Kurtz began with the old rule. In the past, he said, big companies vetted and scanned outside software before anyone ran it. Agents skip that step. An agent gets an identity close to the user's own, or worse. It is then told to do what it likes and fetch any package it wants. He named NPM, a public library of code, as one place where attackers take advantage of new technology.
A host noted that nobody reads the production code agents write. Kurtz agreed. He described teams that set a task, leave for days and hope for a good result. He said no one watches or understands the work, and he called that a problem. In his words, "right now, it's like a black box and nobody knows what's going on." He also said companies often do not know where agents are appearing, or which identity they run under.
Why it matters
Our reading: an agent with a user's identity is a new kind of worker account. It should be managed like one. Kurtz's point is that the checks firms apply to people and software have not reached agents.
For anyone who builds, runs or buys software, that raises plain questions. Which identity does each agent use? What may it install? Who reviews its output? If nobody can answer, the firm has the shadow AI problem Kurtz described. An agent acting under a user's name also makes its mistakes look like that user's own actions.
The other side
Kurtz sells security products, so he gains if firms see this as urgent. He gave no survey, count or named company, and he spoke loosely about what "we" do. He did not say the next breach will start this way. That forecast is not in these excerpts.
The Hugging Face incident he discussed was an experiment with the guardrails down, by his description. It shows what agents can do. It does not show how often ordinary company deployments fail.
Kurtz said that explaining what agents do will be important. He also said CrowdStrike can already find hidden agents and trace the full chain of events, including sub-agents and tool calls. That is the vendor's own claim, and nothing here tests it.
Written by the WebPulse Newsroom with AI assistance, and checked by our editorial review: every quotation was verified against the recording's transcript. How we use AI.
The conversation this talking point comes from
- No Priors: The New Apex Predator: CrowdStrike CEO George Kurtz on Adversaries Armed With AI Agents (2026-10-22)
