Blog

August 2026 · 8 min read

Ghostjacking

Ghostjacking uses indirect prompt injection to turn trusted AI agents into attack tools. Learn how the attack works and how runtime enforcement stops it.

Read the research
April 2026 · 8 min read

Hidden Channel

This isn't a theoretical attack. It's what happens, by default, any time a developer signs into Claude on both a corporate and a personal device.

Read the research