Latent Minds works on adaptive intelligence across learning, interpretability, and alignment.
Research
How models represent themselves, other agents, and their objectives.
More capable models can plan across longer horizons and act with less supervision. We study the internal representations and learned strategies that make those behaviours possible.
Our questions include when self models emerge, how cooperation changes under pressure, and which mechanisms remain stable enough to evaluate or steer.
Explore the researchBiosafety
Safeguards should be measured under adversarial use.
Genome and protein models can lower the cost of biological design. We try to establish how much attacker effort their safeguards will withstand.
We compile the public evidence and design evaluations of safeguard durability, attacker cost, and failure modes.
Read the biosafety evidenceCrucible
Study behaviour inside worlds.
Static benchmarks capture a model at one moment. Many safety failures develop across time, incentives, other agents, and the consequences of earlier actions.
Crucible is our programme for controlled, long horizon reinforcement learning environments. We are building environments where cooperation, negotiation, deception, and moral tradeoffs can be observed and repeated.
Enter CrucibleLearning
The smartest minds should be working on safety.
Advanced AI may shape scientific work, institutions, and the distribution of power. Keeping these systems steerable is not a problem for one discipline or one lab.
We are building pathways into the technical literature and broader awareness of the field so people from machine learning, biology, neuroscience, mathematics, security, and governance can test the arguments and move toward original work.
Start learningJoin the work
We're a growing team making an outsized impact.
We are looking for people who care about the future of humanity.
Work with us