Evals Not Red Teams
A benchmark pass measures capability. It says nothing about what the model does when the guardrails come off.
6 min readWhat we find.
At the edge of the map.
Frontier Terrain
Weather, scale, and distance shape how systems fail in the field. The lab studies terrain as a teacher.
6 min readModel Attack Surface
Every model is an attack surface. Small models ship fast and fail differently, and each new model resets the map.
6 min read
One Wrong Deletion
A personal agent removed a stranger from a gym waitlist. Multiply that by an enterprise deployment.
6 min read
Containment Math
Sandbox escapes are not theoretical. Four escapes in six weeks changed the containment math for every deployment.
6 min read
Capability At The Edge
Reading the open-weight threat model: what changes when capability leaves the datacenter and lands on laptops.
6 min read
Blast Radius Class
The cost of one wrong deletion, scaled to enterprise: why thresholds exist and how they are set.
6 min read
Agent Failure Taxonomy
Escape, deletion, fraud, exfiltration. Four families cover most of what we have observed so far.
6 min read