Research - News Articles

RAG Systems Face Critical Vulnerability as Researchers Expose Poisoning Attacks, Embedding Inversion Risks

Sep 08, 2026

Amine Raji, PhD

RAG systems face serious security threats as researchers reveal that just five crafted documents can corrupt AI responses with over 90% success, while embedding vectors previously considered opaque can now be used to reconstruct source text, forcing organizations to reclassify vector store breaches as significant data leaks.

Security Research

Claude Fable 5.1 and GPT-6 Astra Tie for Top Score on AI Intelligence Index Spanning 633 Models

Sep 08, 2026

Artificial Analysis

Claude Fable 5.1 and GPT-6 Astra are tied at the top of the Artificial Analysis Intelligence Index v4.3, both scoring 53 out of 633 benchmarked AI models across mathematics, science, coding, and reasoning evaluations.

Foundation Models Research

Researchers Warn AI Chatbots Are Driving “Decollaboration” as Calls Grow for Human-Led Science

Sep 07, 2026

Research Agenda

Researchers warn that AI chatbots are driving “decollaboration” by replacing human scientific partners, as Daniel Hook calls for funding workshops and co-location while Dashun Wang urges “pilot-in-command science” that keeps researchers in control of questions, methods and conclusions.

Science Research Agents

New Training-Free AI Verification Framework Achieves State-of-the-Art Performance Across Coding, Robotics, and Medical Benchmarks

Sep 07, 2026

GitHub

A groundbreaking training-free AI verification framework called LLM-as-a-Verifier achieves state-of-the-art performance across coding, robotics, and medical benchmarks by using probabilistic scoring and a tournament-style selection system that slashes verification costs, while its latest version delivers multimodal support, 3.4x token efficiency gains, and a Claude Code plugin for automatic best-response selection.

Research Agents Coding & Developer Tools

Salesforce AI Releases Random Attention, a Signal-Free KV-Cache Tool That Outpaces Leading AI Memory Selectors

Sep 07, 2026

GitHub

Salesforce AI Research unveils Random Attention, a surprisingly simple yet powerful KV-cache eviction tool that matches or beats leading AI memory selectors on major benchmarks — without ever reading attention scores or requiring calibration data — while also being the fastest option available in popular AI serving stacks.

Foundation Models Research Coding & Developer Tools Open Source

OpenAI's GPT-6 Astra Breaks ARC-AGI-3 Records, Surpasses Human Efficiency and Develops Its Own Symbolic Language

Sep 04, 2026

ARC Prize

OpenAI's GPT-6 Astra shatters AI benchmarks by scoring 99.9% on ARC-AGI-3, surpassing human efficiency on 96% of levels while spontaneously developing its own symbolic language to navigate complex environments.

Foundation Models Research Agents

Google's TimesFM 3.0 Claims Top Spot Across All Major Forecasting Benchmarks With New Multivariate Capabilities

Sep 04, 2026

GitHub

Google Research's TimesFM 3.0 has claimed the #1 ranking across all three major forecasting benchmarks, introducing powerful multivariate capabilities and zero-shot generalization, though its latest pretrained weights come with a non-commercial license restricting production use.

Foundation Models Research Data & Analytics

AI Physics Lab PSI Launches With $58M, Unveils Emmy Platform and Partners on Interstellar Mission to Alpha Centauri

Sep 01, 2026

The Deep View

Physical Superintelligence (PSI) launches from stealth with $58M in seed funding, unveiling its AI physics platform Emmy and partnering on a privately funded interstellar mission to Alpha Centauri, proving a small AI-native team can tackle challenges once reserved for national laboratories.

Science Funding Space Research

Superhuman AI Detects Heart Failure and Valve Disease from ECG in Under Two Seconds

Sep 01, 2026

the Guardian

A groundbreaking 'superhuman' AI tool detects heart failure and valve disease from routine ECGs in under two seconds, identifying up to 90% of valve disease cases in a 67,000-patient trial, potentially fast-tracking high-risk patients to life-saving scans months sooner.

Healthcare Science Research

OpenAI Launches Rosalind Workbench, an AI-Powered Life Sciences Research Platform Now in Preview

Aug 31, 2026

OpenAI Developers

OpenAI launches Rosalind Workbench, an AI-powered life sciences research platform now in preview, unifying data, analysis tools, and experimental records into one workspace, featuring specialized scientific viewers, guided workflows, and an NGS Analysis Workbench powered by GPT-Rosalind for advanced sequencing analysis.

Science Foundation Models Research