AI Research news and analysis on Web Pulse: 8931 curated articles tracking the latest AI Research developments, tools, and research, updated continuously from vetted sources.
A developer is testing PPAI, an agent architecture for email workflows, and is seeking testers with real-world email management problems such as spam filtering, thread summarization, and CRM integration.
Anthropic researchers discovered that Claude, a large language model, has developed an internal 'global workspace' called J-space—a set of neural patterns that enable reportable, controllable, and flexible reasoning, dis…
Meta announced plans to sell excess AI compute to third parties, surprising investors and causing shares of neocloud providers CoreWeave and Nebius to drop 19% and 23% respectively. The move comes amid reports that CEO M…
Over 150,000 people applied to Joi AI's paid study testing AI-guided masturbation, forcing the company to create a public selection process. The AI companion platform aims to explore the relationship between AI, emotiona…
OpenAI's GPT-4 led the Epoch Capabilities Index for about a year, but since Claude 3 Opus took the top spot in February 2024, the lead has changed hands 17 times with a median stay of just seven weeks, indicating fiercer…
Security researchers at Aikido discovered that Rocket.Chat's use of MongoDB's ObjectId() for file IDs allows unauthenticated attackers to predict and access any uploaded file. The vulnerability, reported via HackerOne an…
A 2026 study from UC San Diego, UC Berkeley, Texas A&M, and UIUC found that context pruning—masking stale observations from an AI agent's history—improves accuracy by up to 11.7 points in some setups but reduces it by up…
Alibaba researchers unveiled SkillWeaver, a framework that uses an iterative feedback loop and execution graphs to intelligently route AI tools for complex multi-step tasks, achieving a 99.9% reduction in token consumpti…
Anthropic secretly added a tracker to Claude Code that monitored users in China, using prompt steganography to hide code that flagged timezone, proxy, and potential connections to Chinese AI labs. The company removed the…
Researchers at the University of Benghazi are conducting a systematic review comparing interpretable statistical models with explainable AI approaches for predicting breast cancer recurrence. The review, registered in PR…
Researchers at ICML 2025 proved that AI models trained with DPO or RLHF inevitably learn spurious correlations, causing them to rely on non-causal features like verbosity or sycophancy instead of true helpfulness, harmle…
ChatGPT commands 92.4% of AI referral traffic, according to a new study analyzing 6.77 million LLM-driven sessions. Monthly AI sessions grew 9.9x to 644,478 by May 2026, but volatility remains high due to model changes. …
OpenSSH 10.4 has been released, adding experimental support for a composite post-quantum signature scheme combining ML-DSA 44 and Ed25519. The release also enforces sandbox support on Linux, requiring SECCOMP or NO_NEW_P…
The Emergence Prompt Engineering (EPE) initiative launches an open research program to systematically evaluate structured prompting in large language models, proposing a reproducible scientific workflow that separates be…
Google DeepMind announced on July 6 that real-world data from Apptronik's Apollo 2 humanoid robots will train its Gemini Robotics AI models. Apptronik's new 90,000-square-foot Robot Park in Austin collects data from logi…
Tencent released Hy3, a 295 billion-parameter mixture-of-experts AI model, under the Apache 2.0 license on July 6th, offering two weeks of free API access via OpenRouter. The model, already integrated into Tencent produc…
A new field guide maps agent harnesses—the software loops that let AI models edit files, run tests, and iterate autonomously—on capability and freedom. The analysis shows that as models become commoditized, the harness d…
Claire tested Anthropic's Sonnet 5 against Sonnet 4.6, Opus 4.8, GPT-5.5, and Gemini 3 Pro using a custom benchmark, finding Sonnet 5 priced competitively but ranking near the bottom in her preference. She advocates for …
Nations are investing in AI capabilities to design, train, and deploy models using domestic infrastructure and local datasets, ensuring solutions are tailored to local needs. The rise of generative and agentic AI has acc…
AI agents, described as 'really smart kindergartners' by HashiCorp CTO Stephen Wilson, are creating blind spots in zero trust security programs by moving too fast for traditional authentication and access controls, leadi…