Google Research
·
3 months ago
Two AI agent frameworks were introduced to assist academic researchers: PaperVizAgent generates publication-ready figures from manuscript text using a five-agent collaborative system, while ScholarPeer automates peer review by dynamically searching literature and verifying technical claims. PaperVizAgent achieved a score of 60.2 on a 0-100 scale, exceeding the human baseline of 50.0 and outperforming baselines like GPT-Image-1.5 and Paper2Any. These tools aim to reduce administrative overhead in research workflows and improve the quality and efficiency of academic publishing, though both are experimental prototypes not yet production-ready.
Amazon Science
·
3 months ago
Amazon built RuleForge, an agentic AI system that automatically generates security detection rules from vulnerability exploit code. The system achieved 336% faster rule creation compared to manual methods over the final four months of 2025 while maintaining production-grade accuracy. This automation allows security teams to translate vulnerability disclosures into protective detection logic at scale, significantly reducing the time gap between when vulnerabilities are publicly disclosed and when defenses are deployed.
OpenAI Blog
·
3 months ago
● 3 sources
OpenAI released new enterprise AI products including Frontier, ChatGPT Enterprise, Codex, and company-wide AI agents as business adoption of AI systems accelerates. The tools are designed to handle increasingly complex enterprise workflows across multiple departments simultaneously. Organizations can now deploy AI agents that operate independently across their entire operations rather than in isolated functions.
OpenAI Blog
·
3 months ago
OpenAI released a Child Safety Blueprint outlining principles for developing AI systems that protect minors online. The document provides recommendations across safeguards, age-appropriate design, and industry collaboration, though no specific implementation timeline or measurable metrics were disclosed. Organizations building AI systems are expected to use these guidelines to reduce risks of child exploitation and harmful content exposure.
Hugging Face Blog
·
3 months ago
Safetensors, a file format for storing machine learning model weights that prevents arbitrary code execution, has joined the PyTorch Foundation as a Linux Foundation-hosted project. The format is now used as the default for model distribution across Hugging Face Hub and tens of thousands of models across all modalities. The transfer moves governance of the project from Hugging Face to the Linux Foundation, establishing vendor-neutral stewardship while the format and APIs remain unchanged for existing users.
