Google Research Debuts New Statistical Framework for Auditing Machine Unlearning

Google

Jun 15, 2026

Google Research introduced Regularized f-Divergence Kernel Tests, a statistical framework for auditing machine unlearning and differential privacy. The method detects localized data shifts with higher sensitivity and fewer samples than traditional tools. It uses adaptive f-divergences to eliminate manual hyperparameter tuning, successfully identifying privacy violations in mechanisms that previously required millions of samples to detect.

View the full update on research.google

Google Research

@GoogleResearch4d ago

Presenting Regularized f-Divergence Kernel Tests, a new framework for auditing machine unlearning and differential privacy. This method is more sensitive to localized data shifts and requires fewer samples than traditional tools. Learn more: https://t.co/FdPcfJeEs4 https://t.co/MHxy5TdAJE

21181

View on X

Every HeadsUpAI update is written based on its original source and reviewed before it's published. Read our editorial standards →

Keep reading

Anthropic Develops Diff Tool to Surface Hidden Behavioral Differences Between AI Models

Anthropic researchers introduced the Dedicated Feature Crosscoder, a tool that identifies unique internal features across different AI model architectures. By isolating model-exclusive behaviors like political bias or refusal mechanisms, auditors can proactively discover unknown risks that traditional benchmarks miss.

Google DeepMind Releases Toolkit to Measure How AI Manipulates Human Behavior

Google DeepMindMar 28

Google DeepMind Releases Toolkit to Measure How AI Manipulates Human Behavior

Google DeepMind released a new evaluation framework and study of 10,000 participants to measure how AI models can harmfully manipulate human decision-making. The research identifies specific tactics like fear-mongering and establishes a toolkit to track a model's propensity to exploit emotional vulnerabilities.

Google Research Boosts RAG Accuracy with Iterative Agentic Context Search

GoogleJun 7

Google Research Boosts RAG Accuracy with Iterative Agentic Context Search

Google Research and Google Cloud introduced a new agentic RAG framework designed to handle complex enterprise queries. This framework employs a multi-agent workflow that iteratively searches for sufficient context, improving accuracy beyond standard Retrieval-Augmented Generation (RAG). It aims to deliver more dependable responses by preventing the AI from guessing when information is incomplete across multiple data sources.

NotebookLMJun 7

NotebookLM Reveals AI Generation 'Formula' for User Control

Google's NotebookLM has launched a new Source Attribution feature. This update provides visibility into the exact prompts and source materials that generate AI outputs, enhancing transparency and control over the content. It also enables customization of these underlying "formulas" to refine AI-generated artifacts.