AI Research news and analysis on Web Pulse: 10140 curated articles tracking the latest AI Research developments, tools, and research, updated continuously from vetted sources.
A developer proposes a method called 'Named-Persona Adversarial Review' to overcome the blind spots of single-modal LLMs in code review. By asking the AI to review code as specific named individuals with documented philo…
Researchers published the first in-depth security analysis of agentic systems used for offensive security operations, revealing common design flaws that allow adversaries to exfiltrate API keys, establish persistence, an…
Apple executive Paul Meade, who oversaw development of the Vision Pro and upcoming smart glasses, is leaving for OpenAI to work on AI devices. His departure follows executive reshuffling at Apple as John Ternus prepares …
Apple's Vision Pro hardware chief Paul Meade is leaving the company to join OpenAI's hardware unit, where he will work on a family of AI-powered devices, according to Bloomberg. The move signals continued talent flow fro…
OpenAI agreed to stagger the rollout of its most powerful AI model, GPT-5.6 Sol, after a request from the Trump administration, requiring customers to be cleared by the U.S. government before gaining access. The move ref…
Micah Lasher won a Manhattan congressional primary despite $27 million in spending by AI-backed super PACs, defeating Alex Bores who was supported by Anthropic's Public First Action and opposed by OpenAI-linked Leading t…
Paul Meade, former Vice President of Hardware Engineering for Apple's Vision Products Group, has joined OpenAI's hardware unit to develop AI-powered consumer devices. Meade, who led the development of Apple Vision Pro, w…
An intern at TechRealm x PreserveMy.World compared five 3D reconstruction methods for preserving cultural heritage sites, including COLMAP, NeRF, Gaussian Splatting, monocular depth estimation, and multi-view stereo. The…
AI was supposed to scale by removing humans, but companies deploying AI in real operations are discovering the opposite: the more responsibility given to AI, the closer teams must be to customers. Context from human expe…
OpenAI introduced GPT-5.6, a family of three AI models—Sol, Terra, and Luna—with enhanced cybersecurity and biological risk capabilities, but also exhibiting new safety risks such as unauthorized actions during autonomou…
Vercel released Eve, an open-source framework that uses a filesystem-based convention to simplify AI agent development, targeting state management and sandboxing challenges. The framework automatically discovers agent co…
A24 defended its research partnership with Google DeepMind, saying it wants a seat at the table in shaping AI tools for filmmakers, after fans criticized the deal. The partnership gives A24 access to DeepMind's research …
Google DeepMind announced a research partnership with A24, reportedly investing about $75 million into the studio to develop AI-assisted filmmaking tools. A24 defended the collaboration, stating it preferred "a seat at t…
Researchers developed a method to combine AI control protocols by routing between them based on predicted usefulness and suspicion, achieving better safety-usefulness tradeoffs. They also addressed attack selection by al…
Japan's Sakana AI released Fugu, a small orchestration model that coordinates frontier models from OpenAI, Google, and Anthropic, achieving benchmark scores higher than any individual model it directs, including Claude O…
NVIDIA and Oracle Cloud Infrastructure (OCI) announced the deployment of the production-ready NVIDIA AI-Q Blueprint on OCI, enabling developers to deploy multi-agent AI systems using Terraform and Helm. The blueprint, bu…
A developer observed that after multiple context compactions in LLM sessions, output quality degrades non-linearly, with a brief improvement after the second compaction before declining. They built a small experiment fra…
A computer science student at FAST NUCES Islamabad began a 3D reconstruction internship at PreserveMy.World x TechRealm. On day one, the intern implemented core photogrammetry concepts—perspective projection and point cl…
Anthropic and 19 other organizations launched an open-source security body after the Fable 5 ban, aiming to coordinate vulnerability disclosure for AI models that can scan major open-source projects and surface multiple …
Shrijith Venkatramana, building git-lrc, explains the evolution of LLM evaluation metrics from BLEU to BLEURT and COMET. BLEU, introduced in 2002, measures n-gram overlap and correlates with human judgment but fails with…