"Every enterprise has a firewall protecting its network. Now they need one protecting their AI agents."

Robots

People

Love

Reading time: 3 min.

A recent survey of 202 enterprise security leaders by Enterprise Management Associates finds 65% of organizations have already had an AI agent act outside its intended scope: reaching systems it wasn't cleared for, sharing data it shouldn't, or exposing credentials it was never meant to hold.

94% of leaders say they're confident their access controls hold up. Only 33% actually provision agents with least-privilege access, and just 34% check an agent's authorization at the moment it acts, not only at setup.

What the disciplined third do differently:

94% say their controls hold up. 33% actually enforce them.

How human-centric AI delivers business results

Your weekly briefing on tools that create competitive leverage

Discovers every AI agent in a company and blocks any skill, plugin, or MCP server that fails a security check. $50M seed (Sequoia, then Greenoaks), out of stealth September 1. Best for CIOs who can't say what every agent is allowed to touch.

Flags any part that deviates from known-good, no defect list required, checking X-ray, CT, and sensor data in real time. Launched September 3, built on the first foundation model trained at scale on industrial CT data. Best for manufacturing leaders who find defects after they've already shipped.

Hendricks co-founded AGent Energy after clocking how much backup generation sits idle at hospitals, factories, and universities, more spare power than most utilities can call on when the grid is stressed. Her AI platform aggregates and dispatches that behind-the-meter capacity, targeting more than 200 gigawatts nationwide. AGent closed an $11M Series Seed on August 13, 2026, led by Spero Ventures and MassMutual Ventures.

I keep seeing the same mistake with agent rollouts: a team gives an agent access to a system, and nobody writes down what it's actually allowed to do there. LEAP starts before the access does. Locate the decision the agent will make. Evaluate what happens if it makes the wrong one. Build an action plan: the scope, the approval gate, the kill switch. Then track progress against that plan, not just uptime.

Twelve-week cycles work the same way. Write the scope down in week one, or you're auditing a system nobody agreed to build.

Get the full framework in the LEAP Guide.

One book, handpicked from my conversations with friends, industry leaders, and tech innovators:

The Checklist Manifesto by Atul Gawande shows how surgeons, pilots, and builders cut fatal errors using a short list, checked every time, no matter how senior you are. The same logic applies to an agent with system access: write the checklist before the first run, not after the first incident. Perfect for: Gamora-CIO and Gamora-AI who need a scope-and-check discipline that survives past the first agent's launch.

Lighthouse takes us many hours to create, and we don’t run ads or charge anything. We are dedicated to making it the best AI transformation newsletter possible.

We just ask for one thing: tell one friend about Lighthouse today!