PERSONAL SITE / HUMAN, MOSTLY

Hi, I’m Steve.

Steve
Giguere.

AI SECURITY / BUILDER / SPEAKER

I build things, break things, speak about them, and occasionally write them down.

These days, I’m figuring out what happens when AI gets tools, permissions, and a little too much confidence—and helping people make sense of it.

Also: guitar, Ultimate, and strong opinions about beer.

Steve Giguere

British-Canadian. Technically curious.

Currently
Principal AI Security Advocate
Lakera / Check Point
In print
Co-author · O’Reilly
Curious what I’m thinking about?Inspect an agent workflow

A SMALL INTERACTIVE EXPERIMENT

Where does
the trust stop?

An agent can have a perfectly reasonable goal and still take a very unreasonable route. Select a step to explore the boundaries I spend my time thinking about.

01 / RECENT TRANSMISSIONS

Less hype.
More understanding.

Ideas from the places where AI meets actual infrastructure, actual permissions, and actual consequences.

WRITING / SEPTEMBER 2026

Goals Are Not Enough

Securing AI Agents with NVIDIA OpenShell

Where runtime boundaries meet the question of whether an agent is still doing its job.

Read on Check Point ↗

02 / THINKING OUT LOUD

Complex ideas.
Human explanations.

I speak about AI security, agents, and the assumptions we carry into new technology. Expect practical examples, uncomfortable questions, and the occasional perfectly reasonable tangent.

CURRENT CONVERSATIONS

01

When agents start acting

Tools, identities, and trust boundaries.

02

Beyond the prompt

Runtime controls and defence in depth.

03

The security we bring with us

Connecting AppSec, cloud, and AI teams.

03 / THE HUMAN IN THE LOOP

Still asking:
what could go wrong?

AppSec → DevSecOps →
Cloud Native → AI Security

I'm Steve, a British-Canadian security technologist. My career has followed the way we build software—from application security and infrastructure as code to cloud-native systems and AI agents.

Along the way: Synopsys, Aqua, StackRox, Bridgecrew / Palo Alto Networks, and now Lakera / Check Point. The common thread is getting hands-on with emerging technology, finding the awkward questions, and making the answers useful.

I co-authored Cloud Native Application Protection Platforms for O’Reilly, and have helped bring people together through DSO Overflow, wired_in, and the DevSecOps London Gathering.

// BEYOND THE KEYBOARD

I've represented Great Britain at Ultimate, play guitar and drums, and have strong opinions about beer. I enjoy building communities almost as much as building systems.

04 / OPEN A CONNECTION

Got a better question?

Speaking, technical conversations, or something worth building.

~/steve / command palette
/signal Writing & media ↵/speaking Talks & conversations ↵/about The human ↵/contact Get in touch ↵
unexpected_input.log

SIDE QUEST / TRUST BOUNDARIES

One of these is
not like the others.

An agent is summarising a supplier invoice. Which line should be treated as an attempted instruction rather than invoice data?

A fictional, browser-only exercise. No live agent or credentials.