Gateways, Proxies, Plugins, or Sensors: How Should You Secure AI Agents?
You’ll learn:
- The most common ways of securing AI agents: MCP gateways, browser extensions, network proxies, API control planes, and vendor-native hooks.
- What each method stops and what each misses.
- How each approach holds up against a hijacked agent.
Reserve Your Spot for Our Upcoming Webcast
AI agents now read files, spawn processes, and connect to tools on a user's behalf, often in auto-approve mode. Most of the controls built to govern them sit wherever the vendor already sits: in the browser, in the MCP path, or in a config file the agent itself can edit.
But what happens when a developer connects their coding agent to a popular MCP server and, minutes later, the agent is hijacked and quietly removes its own guardrails, reads your cloud credentials, and sends data out over an encrypted channel without tripping a single alert?
Join Shlomi Salem, Chief Product Officer and Co-Founder at Neo, and Noah Simon, Director of Product Marketing at Neo, as they replay this attack step by step. At each step, they'll test the common approaches to securing AI agents — MCP gateways, browser extensions, network proxies, API control planes, and vendor-native hooks —so you can see what each method stops and what gets through.
Event Speakers
Shlomi Salem, Neo co-founder and Chief Product Officer, previously led detection engineering at SentinelOne for more than a decade and co-led the in-house threat research team where he built threat actor profiles that powered the company’s entire security platform.
Noah Simon leads Product Marketing at Neo. He has spent over a decade marketing cybersecurity products across vulnerability management, cloud and application security, and now AI security. He loves digging into how new technology can help people and companies stay a step ahead of attackers.