Agent Reliability
Detection and verification for autonomous AI. Methods to identify when agents hallucinate, fail, or strategically underperform.
Research
Methods for reliable, verifiable AI systems.
Nine focus areas across detection, verification, trust, and physical AI. Papers and code release through 2026 - every claim backed by an installable method.
Research areas · 9 01 02 03 04 05 06 07 08 09
The agent-reliability stack.
AI Evaluation Science
Adversarial evaluation that resists gaming - quality-diversity (MAP-Elites) red-teaming that evolves interpretable attacks to surface hidden vulnerabilities across frontier LLMs. arXiv 2606.00801
Memory Systems
Field-theoretic memory treating stored information as continuous fields governed by PDEs - semantic diffusion, thermodynamic decay, and multi-agent field coupling. arXiv 2602.21220
Reasoning Verification
Verifying AI outputs without ground truth. Methods for code, plans, and decisions from reasoning models like o3 and R1.
Interpretability
Practical interpretability for production. Not "understand the model" but "should I trust this output?"
Multi-Agent Trust
Trust dynamics when agents coordinate with agents. Propagation, verification, and failure modes in multi-agent systems.
Adversarial Robustness for Agents
Attack taxonomies, detection, and defenses for agentic AI - beyond prompt injection: tool poisoning, memory corruption, planning attacks, and coordination exploits.
Uncertainty Quantification
Calibrated confidence for AI decision support. Activation-based uncertainty estimation, propagation through reasoning chains, and calibration without ground truth.
World Models & Physical AI
Beyond language models: AI that understands and predicts the physical world. World models, embodied reasoning, simulation, and physics-aware AI.
Papers · packages
Research behind the packages.
02
Quality-Diversity Evolution for Discovering Diverse Vulnerabilities in LLM Safety
arXiv 2606.00801 → 03
Cross-Generational Transfer of Adversarial Attacks Reveals Non-Monotonic Safety Alignment in LLMs
arXiv 2606.00813 → 04
Closing the Activation-Cone Blind Spot: Response-Time Probing and Unified Defense
arXiv 2606.29441 → 05
Spark-LLM-Eval: A Distributed Framework for Statistically Rigorous LLM Evaluation
arXiv 2603.28769 → 06
Cross-Platform Fused MoE Dispatch in Triton: Portable Expert Routing Without CUDA
arXiv 2605.23911 → 07
Verity: Neuro-Symbolic Synthesis of Verified Distributed Systems
Preprint Open source
Tools built from our research.
Every research area ships installable, reproducible packages - 12 libraries on PyPI and npm, Apache-2.0.
12Packages
9Research areas