Writing

Notes, essays, and research from the lab. Infrequent on purpose.

∎latest

2026 · 04 · 29research · 22 min
Loss of Control: The AI Apocalypse Is Closer Than You Think
Self-preservation as emergent behavior in frontier LLMs — measured across 10 models and 300 runs under termination pressure.
2026 · 04 · 19essay · 6 min
A New Focus
We spent a year on AI security and safety under one driving belief: agents should behave more deterministically. Why that belief led us to stop constraining models and start building the environments they run in.

◈past research

pre-rlvr · 1 report

click to expand

Work from before the lab's pivot to RLVR environments.

2025 · 05 · 19research · 8 min
The Hidden Dangers of Browsing AI Agents
First end-to-end threat model for LLM-based browsing agents — systemic vulnerabilities across planner, tools, and content pipelines, with a defense-in-depth strategy.