AI Ethics news and analysis on Web Pulse: 16832 curated articles tracking the latest AI Ethics developments, tools, and research, updated continuously from vetted sources.
A developer argues that AI agents must never be allowed to choose their own acting subject, as tool arguments generated by a model are untrusted and cannot establish identity or delegation. The post outlines a design whe…
A new analysis of AI governance models finds that global frameworks including the EU AI Act and the U.S. NIST AI Risk Management Framework favor distributed operational accountability, placing final authority over high-i…
A new safety-constrained large language model (LLM) for maternal and child health (MCH) resources, developed by an unnamed team, achieved an average response time of 5.3 seconds in a public health deployment, using a dom…
By 2030, 59 out of every 100 workers will need to reskill or upskill to stay relevant, according to a report on AI-driven upskilling frameworks. The US National Association of State Boards of Accountancy has approved an …
HealthClaw, an open-source AI health coach, boosted answer accuracy from 0.2% to 45.7% on a synthetic year-long benchmark while reducing unnecessary data exposure by 71.7%, according to its developers. The system, design…
The AIMO Interpretability Challenge, organized by the AI Mathematical Olympiad (AIMO) and the Fields Model Initiative, aims to uncover the reasoning processes behind AI models' decision-making by testing them on olympiad…
A new benchmark called Safe-Psych, comprising over 1,000 real-world psychiatric clinical notes, reveals that large language models misdiagnose or overly abstain in over 60% of cases when handling diagnostic uncertainty. …
Elon Musk quietly acquired APR Energy, a medium-sized fossil fuel energy company that provides mobile gas-powered turbines, in May 2026, according to FTC documents and a report by Electrek. The acquisition, estimated at …
A researcher and PhD candidate in tool-making argues that the common phrase 'AI is just a tool — it matters how you use it' is naive and dismissive, pointing out that tools have systemic impacts on environment, policy, a…
Experiments involving over 3,000 participants show that access to AI advice drastically reduces people's willingness to admit uncertainty, with participants three times more likely to answer correctly without AI despite …
Private equity firms are aggressively acquiring ecommerce and payments companies, with Stripe and Advent jointly bidding $53 billion for PayPal, Criteo receiving a buyout offer at a 50% premium, and Klarna raising $517 m…
BitMind Forensics (BMF), trained through Bittensor SN34's open adversarial competition, achieves 0.936 AUC on Sumsub's original images and 0.872 pooled AUC over its full manipulation battery, outperforming state-of-the-a…
Researchers present OriginBlame (ob), a record- and token-level data provenance system that enables precise identification of training records belonging to a specific author, addressing the gap between data contributor r…
A new mathematical framework for underwriting, pricing, and contract design for agentic AI deployments has been developed in a paper on arXiv (2607.13230v1). The framework models risk states including autonomy level, ope…
A new study from researchers at arXiv introduces the Visual Dependency Gap (VDG) metric to audit whether video large language models (LLMs) rely on visual information for correct answers. Testing twenty models from 2B to…
A new framework quantifies the geometric information discarded by machine learning models under Lie group symmetries, enabling applications in data masking, model fingerprinting, and privacy-preserving computation. The m…
A systematic review of Federated Explainable Artificial Intelligence (FedXAI) finds that explainability is transitioning from a post-hoc tool to an integral component of the federated learning lifecycle, according to a n…
A study published on arXiv found that access to AI advice, even when wrong, nearly eliminated participants' willingness to say "I don't know" in five experiments with 3,132 participants. Participants answered more questi…
An AI-generated GPU kernel ran 18.71× faster than an optimized PyTorch baseline, with Fable 5 outperforming other models like Claude Opus 4.8 and GPT-5.5. The achievement raises questions about code review and engineerin…
Linus Torvalds discussed LLM usage in kernel development, but the full content is blocked by Anubis, a proof-of-work system designed to deter AI companies from scraping websites. The administrator implemented Anubis to p…