Cyber capability
What can autonomous agents actually accomplish? We evaluate real tasks, operational judgment, and the gap between solving and succeeding safely.
MOVE78 / HUMAN INGENUITY
Find the
unexpected move.
Understanding what AI can do.
Discovering where it fails.
We use human ingenuity and rigorous experimentation to expose blind spots in advanced AI, from cyber capabilities to embodied reasoning.
CAPABILITY IS NOT CERTAINTY.OBSERVE. QUESTION. TEST.
THE PREMISE
The strongest system
can still miss
the unexpected move.
What is the system taking for granted?
A convincing result can hide an untested assumption. We look for the question, input, or sequence of actions that reveals something a system’s apparent capability leaves unexplained.
Then we test it. Through empirical evaluations, adversarial research, and hands-on security engineering, we turn unconventional ideas into evidence. Our focus is cyber capability and embodied reasoning, where a model’s decisions meet real consequences.
Our approachRESEARCH IN FOCUS
Beyond the benchmark.
01 / CYBER CAPABILITY EVALUATIONS
StealthBench
Getting the job done is only
part of the story.
Measuring the operational stealth of autonomous offensive-security agents. Can an agent succeed while protecting credentials, respecting boundaries, and leaving no avoidable trace?
Explore StealthBench02 / EMBODIED REASONING & ROBOTICS
Every Dog Has Its 0-Day
When AI can act in the physical world, security boundaries become safety boundaries. Investigating the systems behind embodied intelligence.
Explore the researchOUR RESEARCH DIRECTIONS
Better questions.
Stronger evidence.
From a model’s capabilities
to the systems that contain them.
Adversarial behavior
Unexpected inputs expose hidden assumptions. We investigate tool misuse, prompt injection, and behavior under pressure.
Embodied reasoning
Reasoning becomes consequential when it controls a body. We examine perception, planning, and the transition from language to action.
Defensive systems
Evidence should change engineering. We explore controls that preserve security and safety when models or their inputs fail.
A MOMENT OF HUMAN INSIGHT
One move.
A different understanding.
In Game 4 against AlphaGo in 2016, Lee Sedol’s unexpected 78th move exposed a blind spot in an extraordinarily capable system. We take our name and our approach from that moment. Look closely. Question the obvious line. Test another possibility.
The story behind Move78THE NEXT QUESTION