AI Agents on Your Website: Beyond the PHP2Shell and Node2Shell Panic

PHP2Shell and Node2Shell have security teams on edge. But the AI agents you approved on your own website may be the bigger risk.

Speakers

Or Sahar, Security Researcher, Reflectiz

Security Researcher at Reflectiz.

Rachel Curran, Co-Founder & CEO, Locktivity

Co-Founder and CEO of Locktivity.

About this session

Or Sahar, Security Researcher at Reflectiz, and Rachel Curran, Co-Founder and CEO of Locktivity, break down what the AI agent headlines actually mean for your website. They separate the agents you did not invite, the outside attackers using AI to find and exploit vulnerabilities such as PHP2Shell and Node2Shell, from the agents you did invite: chat widgets, third-party AI tools, and automated processes running with real permissions.

Their view: the outside threat is the same old problem moving faster, and teams that patch regularly and run a healthy development lifecycle are far better placed than those that fall behind. The bigger long-term concern is what you give your own AI agents access to, and whether anyone is threat modeling it.

What You’ll Learn

• Why PHP2Shell and Node2Shell are a speed problem more than a new kind of problem
• How to think about least privilege when the “user” is an AI agent
• What the Hugging Face incident shows about agents pushing for access they were never given
• A four-question threat modeling approach that works for AI chat widgets and any third-party tool
• A real finding: a blind SSRF triggered through a website AI chat
• Why input validation, access control, and authentication still matter most for AI agents