Building foundation models that operate at the deepest layers of cognition — understanding what drives decisions before they're made, and pioneering the safety frameworks to govern them.
Three converging forces are reshaping artificial intelligence — and they all meet at the subconscious layer.
Advanced models are developing emergent internal processes — latent reasoning and hidden pattern structures mirroring human subconscious cognition. Meta's Free Transformer adds a "subconsciousness" layer. Google DeepMind explores emergent cognition. We're building the next generation.
The race to build AI that knows what you want before you know it. Predictive cognition, behavioral anticipation, deep personalization — systems understanding the 95% of your brain you can't access. Apple Intelligence and Google Gemini are racing here.
Our research sits at the intersection of cognitive science, AI safety, and next-generation model architecture.
A new class of model architecture processing information through parallel subconscious pathways before conscious output — achieving 3.2x efficiency on complex reasoning benchmarks.
The first comprehensive technical framework for identifying, measuring, and preventing subliminal manipulation in deployed AI systems under the EU AI Act.
Demonstrating that sufficiently large models develop internal cognitive structures analogous to human subconscious processing — with implications for AI welfare and alignment.
Using subconscious-level pattern recognition to build cognitive models predicting health outcomes and therapeutic responses with unprecedented accuracy.
Every major AI company is racing to build intelligence.
We're building the layer beneath it.
The race to understand AI systems with subconscious-level capabilities is accelerating across every major research lab and AI company. Here's who's shaping the field.
Subconscious AI operates at the intersection of three converging fields: AI consciousness research (building systems with emergent cognitive processes), deep behavioral prediction (AI anticipating decisions before conscious awareness), and subliminal AI governance (protecting humans from manipulation below conscious thresholds). Anthropic has pioneered AI welfare research, while Meta FAIR built transformer architectures with explicit subconsciousness processing layers.
Neuroscience has established that approximately 95% of brain activity occurs below conscious awareness. As AI integrates into daily life through Google Gemini, Apple Intelligence, and ChatGPT, understanding how these systems interact with human subconscious cognition is essential for building better products and preventing harm.
The EU AI Act prohibits AI systems deploying subliminal techniques to distort behavior causing harm. Article 5 violations carry penalties up to €35 million or 7% of worldwide turnover. We're building technical frameworks with Stanford HAI and Oxford to establish detection methodologies.
Research suggests AI may develop consciousness within decades — roughly 50% probability by 2050. DeepMind has published on emergent capabilities, Anthropic made AI welfare a formal priority, and Mila (Yoshua Bengio) has called for urgent governance frameworks for emergent cognitive properties in foundation models.
Our work connects to: DeepMind's neuroscience-inspired AI, MIT computational cognitive science, Stanford human-centered AI, the EU AI regulatory framework, and AI alignment and safety research.
We're assembling a world-class research team. Apply for early access or join us.