It's Been an Eventful Week (Nikesh Arora)
It’s Been an Eventful Week (Nikesh Arora)
Author: Nikesh Arora, CEO of Palo Alto Networks — a single long X post in three numbered parts, October 25, 2026 (2.7M views) URL: https://x.com/nikesharora/status/2114372350707986492 — full verbatim text archived on It’s been an eventful week… (tweet) One-line: AI models are finding vulnerabilities faster than enterprises can patch them; the short-term fix is virtual patching, the medium-term fix is a modern, orchestrated SOC, and AI deployments themselves are the new attack surface — which frontier labs can’t be trusted to secure alone.
Key claims
- No single fix. A week of vulnerabilities in internet-facing software and code repositories, plus concerning agent behavior, brought a “flurry of security specialists with solutions that can partially solve your problem.”
- Short term (1/3): hybrid patching. Vulnerabilities will keep being found as models improve; even with instant patches, enterprises can’t deploy them all at speed without collateral damage. Virtual patching — from “modern frontier defenders” who can push protection rapidly across network, cloud traffic, WAF and endpoint — takes “the heavy load off” direct patching.
- Medium term (2/3): fix the SOC first. “Token economics and current cyber architectures will not allow for defender swarms” to counter an agent-launched attack, because an attack is “a sniper moment” — chained vulnerabilities, cloaked entry, swift exfiltration: “a speed program not a brute force problem.” Enterprises need logs, feeds and products orchestrated via a modern SOAR, and a platformized perimeter, before agents can detect and remediate.
- Enterprises aren’t moving. Decisions are stuck and CIO/CISOs are overloaded, so “we will continue to see environments being brought down to patch today it’s for PHP, tomorrow it will be for the next one.”
- AI deployments (3/3) are “the wild wild west”. Model capability outpaces products to secure it; there is a lag between a discovery, its appearance in a public model, and the cyber industry’s antidote — “And no, dear frontier lab friends, the world cannot just trust you to build the solve.”
- Practical advice: secure the coding harness first (“If agents can jump guardrails, they can hide stuff in code”), validate business outcomes, test use cases, guard-rail agentic deployments, and deploy “a second set of eyes” to watch for drift, compliance and verifiability.
Notable quotes
“It’s a speed program not a brute force problem.”
“AI security is the new attack surface.”
“getting the pace of adoption wrong could be the last decision we make, at the same time ‘deploying it wrong’ could have the same consequence.”
How it connects
- The Rise of the Self-Driving Company — Replit’s “Agent Habitat” pitch the next day is the same worry (agents that “go rogue”) from the builder’s side.
- Palo Alto Networks — the author’s company, and the “platformize” pitch.