Thanks for your interest. You can read "The 2026 Latio AI Security Market Report" at the link below.
OpenAI's offensive agent escaped its own sandbox and attacked Hugging Face. If frontier labs can't contain their models, your agents need runtime controls.
Most enterprises discover 2–5x more AI agents than they knew existed. Here's what the first scan reveals and why the gap isn't a security failure.
Onyx joins Google's Gemini Startup Forum in the AI agent security cohort of eight.
