1stProtect Research

Why we stopped looking at 'Who' and started looking at 'How'. The physics of data movement tells us more than a login token ever could.

Frontier pacing buys time — but only if we build enforceable boundaries between AI capability and authority. The nearer risk is granting agents more control than we intended.

A flagship report on what enterprise AI agents can actually do — credential reach, production tooling, and unapproved destinations — published only from measured, consented environments.

A practical benchmark of what Claude Code and similar coding agents can reach on a standard developer machine — and which surfaces matter for least privilege.

EDR asks “is this malicious?” Coding agents require “is this action allowed for this actor?” Ten high-impact actions that often pass threat detection unchanged.

A CISO-facing guide to least privilege for autonomous actors: principles, operating model, metrics, and how to avoid governance theater without enforcement.

A printable checklist for scoping and running a fourteen-day Audit Mode exposure assessment on coding agents and tool-using copilots.

Same identity, different actor: which permissions should differ between a developer and Claude Code on the same laptop — and how to enforce the gap at runtime.

No data was stolen. The investigation still costs half a million to $1.3 million. Here's the line-by-line bill for the Hugging Face weekend intrusion on a standard enterprise stack.

Why "Friendly Fire" proves the case for runtime enforcement — and why knowing where your agents are is not a security control.

AI agents are our new coworkers — running in browsers, clouds, and infrastructure. But they are non-deterministic, and the "Shadow AI" crisis they create is one traditional security stacks are fundamentally unequipped to govern. Leadership must pivot from monitoring access to enforcing intent.

The major security vendors have settled on their 2026 story: agentic AI security is the new perimeter. The pivot is real. But underneath the shared vocabulary there's a structural gap in what these platforms actually do. They watch. They don't stop.

Yesterday, Anthropic launched Project Glasswing — a coalition built around a sobering realization: AI models can find and exploit vulnerabilities faster than any human attacker. This is the inflection point 1stProtect was built to address.

Why a Clean Pentest Report Is a Dangerous Illusion

It’s Leaking Your Soul

AI Analytics creates faster autopsies. It does not create survival.