Neural Networks
SymTorch: Extracting Symbolic Math from Neural Networks
New framework converts opaque neural network decisions into interpretable mathematical expressions, enabling better model verification and understanding of AI behavior.
Research
deepfake detection
Study: Humans Outperform AI at Detecting Deepfake Videos
New research reveals a surprising detection gap: while machines excel at spotting deepfake images, humans consistently outperform AI systems when identifying synthetic videos.
AI security
CREDIT: Certified DNN Ownership Against Model Extraction
New research introduces CREDIT, a certified framework for verifying deep neural network ownership and defending against model extraction attacks through provable security guarantees.
AI Agents
MIRA: Combining Memory Systems with RL for Smarter AI Agents
New research introduces MIRA, a framework that integrates memory architectures with reinforcement learning while minimizing expensive LLM calls, advancing efficient autonomous agent design.
AI Agents
2025 AI Agent Index: Mapping Safety Features Across Deployed Syst
New research systematically documents technical and safety features across deployed agentic AI systems, creating a comprehensive index for understanding how autonomous AI operates in the wild.
AI ethics
Mirror: Multi-Agent AI System Automates Ethics Review
New research introduces Mirror, a multi-agent framework using AI to assist in ethics review processes, potentially transforming how AI systems evaluate content for safety and compliance.
AI Agents
MAPLE: New Sub-Agent Architecture for AI Memory and Learning
New research introduces MAPLE, a sub-agent architecture enabling memory, learning, and personalization in agentic AI systems through modular design patterns.
AI security
PBSAI: Multi-Agent Architecture for Enterprise AI Security
New research proposes a multi-agent AI reference architecture for securing enterprise AI deployments, addressing governance challenges in managing AI systems at scale.
Machine Unlearning
Temper-Then-Tilt: A New Framework for AI Model Unlearning
New research introduces a principled approach to removing harmful concepts from generative AI models using tempering and classifier guidance, with major implications for synthetic media safety.
synthetic data
PRISM: Structure-Aware Privacy Budgets for Synthetic Data
New research introduces PRISM, a differentially private synthetic data framework using structure-aware budget allocation to optimize prediction accuracy while maintaining privacy guarantees.
AI safety
Research Reveals Early Fractures in Agentic AI Oversight Models
New research examines how AI communities are splitting on human control approaches for autonomous agents, finding significant divergence in oversight philosophies that could shape the future of AI governance.
LLM
Error-Localized Policy Optimization: A New Approach to LLM Tool R
New research introduces ELPO, a training method that teaches LLMs to learn from irrecoverable errors in tool-integrated reasoning chains, improving agent capabilities.