Assail | Adversarial Exposure Validation for the APIs, Web Apps, Mobile Apps

Our Blog

Research, perspective, and field reports from the team building autonomous offensive security.

OUR BLOG

Where offensive AI meets the real world.

Intelligence from the swarm — research, releases, and field notes from the team building Ares.

Reaper Learns to Debug Himself

We taught our offensive-security agent to do more than fix his mistakes. Now he fixes the part of itself that makes them.

Created by:

Alissa Knight

Reaper Learns to Debug Himself

We taught our offensive-security agent to do more than fix his mistakes. Now he fixes the part of itself that makes them.

Created by:

Alissa Knight

When should an AI say no?

Ask a large language model for malware, attacks, or dangerous instructions, and it will usually refuse—a behavior known as refusal, which is often exactly what we want.

Created by:

Ilir Osmanaj

Mythos Is Not Your Red Team

What the Cybersecurity Industry Is Getting Wrong About Anthropic's New Model

Created by:

Alissa Knight

Nobody Got Breached by a False Positive

The security industry spent a decade worrying about noise. The thing that actually gets you is silence — and QA figured that out a long time ago.

Created by:

Shqiperim Krasniqi

The Harness Wars

The Next Competitive Frontier Isn't Model Size. It's What You Wrap Around It.

Created by:

Alissa Knight

The $6.6 Billion BOLA

What Lovable's 48-Day Exposure Tells Us About the Collapsing Economics of Human Penetration Testing

Created by:

Alissa Knight

Reaper Learns to Debug Himself

We taught our offensive-security agent to do more than fix his mistakes. Now he fixes the part of itself that makes them.

Created by:

Alissa Knight

Nobody Got Breached by a False Positive

The security industry spent a decade worrying about noise. The thing that actually gets you is silence — and QA figured that out a long time ago.

Created by:

Shqiperim Krasniqi

When should an AI say no?

Ask a large language model for malware, attacks, or dangerous instructions, and it will usually refuse—a behavior known as refusal, which is often exactly what we want.

Created by:

Ilir Osmanaj

The Harness Wars

The Next Competitive Frontier Isn't Model Size. It's What You Wrap Around It.

Created by:

Alissa Knight

Mythos Is Not Your Red Team

What the Cybersecurity Industry Is Getting Wrong About Anthropic's New Model

Created by:

Alissa Knight

The $6.6 Billion BOLA

What Lovable's 48-Day Exposure Tells Us About the Collapsing Economics of Human Penetration Testing

Created by:

Alissa Knight

TEAM

Let's Talk

Reach out and one of our team members will respond within 1 business day.