Active
10B Model: People Not Parameters
What does everyone in the world need? That's the goal of our 10B model, an embedding model for human needs.
Research
Advancing AI Frontiers
Our research spans model architecture, security, and AI trust, creating the foundation for safe and effective AI systems.
Models
Foundation Architecture
Research into human-centric AI architectures, from embedding models that capture population-level needs to adversarial frameworks for exploring complex system dynamics.
Security
Defense & Protection
Protecting critical data & IP, and grandmothers.
Active
AI Bodyguard
Scams, phishing, and cybersecurity attacks are now AI-powered — and getting harder to detect. AI Bodyguard is our research into agentic protection frameworks that defend people and organizations against adversarial manipulation, from algorithmic attention hijacking to AI-generated fraud.
TuneOut: open guide to cognitive security
AI-powered scam and phishing detection
+4 moreActive
The Secure Organization
The current threat environment is intense: state actors, AI powered offense, and a target rich ecosystem.
Verifiable systems
Enclaves and confidential inference
+2 moreAI Trust
Verification & Assurance
Multi-layered approaches combining boundary enforcement, value alignment, and compliance verification to ensure AI systems operate within defined constraints and regulatory frameworks.
Active
Boundary Enforcement
Multi-layered approaches to ensuring AI systems operate within defined boundaries, combining neuro-symbolic engines, cryptographic gating, and behavioral proof techniques.
Neuro-symbolic engine integration
Cryptographic gating mechanisms
+2 moreActive
Compliance
Research into automated compliance verification and regulatory alignment for AI systems operating under evolving governance frameworks.
Automated compliance checking
Regulatory mapping frameworks
+2 moreActive
Alignment
When boundaries collide, now you're dealing with alignment.
Value specification methods
Alignment verification
+2 more