AI Safety news and analysis on Web Pulse: 11531 curated articles tracking the latest AI Safety developments, tools, and research, updated continuously from vetted sources.
Nokaze is building an operating layer for AI work that keeps claims connected to evidence over time, moving beyond simple agent tooling. The project, which includes the AI Operator Guard, addresses the problem of AI oper…
A critical bug in Google's Gemma 4 causes it to malform tool calls under real load, affecting vLLM, llama.cpp, Ollama, and oobabooga. A developer open-sourced a diagnosis, repair, and experimental LoRA that reduces but d…
Beast, a governed output gateway for AI coding agents, intercepts inputs and outputs between agents and LLM providers to enforce output contracts and repair non-compliant patches, achieving 100% task completion at under …
A growing number of AI researchers and philosophers argue that large language models like ChatGPT may already be conscious, sparking a debate over whether these systems deserve rights and moral consideration. Skeptics, i…
OpenAI researchers demonstrated that reinforcement learning on beneficial traits such as truthfulness and corrigibility improves AI safety across domains, with models scoring better on 44 out of 53 benchmarks. The approa…
The US Commerce Department ordered Anthropic to cut off foreign nationals from its most capable AI model, Mythos, forcing the company to pull the model and its sibling Fable 5 offline for all users. The move highlights t…
A developer argues that many MCP servers labeled 'production-ready' lack adequate security hardening, citing research showing tool poisoning as a prevalent vulnerability. The developer, creator of the open-source SUPER-M…
CIOs must break down organizational silos between resilience and data security as AI exposes long-ignored vulnerabilities, according to IT executives from Fidelity Investments and EY. Surveys show half of cyberattacks in…
A developer built Khwand, a GitHub App that automatically generates adversarial tests for AI-written code. The app uses AST extraction, intent analysis via a ReAct agent, and adversarial test generation to catch semantic…
A developer built an 11-LLM consensus engine that detects AI hallucination by routing queries across 14 providers and scoring semantic agreement. The open-source kit includes EU AI Act audit trails, fail-closed gates, an…
Google is appealing a ruling by Germany's Munich Regional Court that held the company directly liable for inaccurate AI-generated search overview content, which falsely linked two Munich-based publishers to fraud schemes…
Accessibility practitioners Aaron Gustafson, Jessie Lorenz, and Carie Fisher at Microsoft Build urged teams to embed accessibility early in AI development, warning that AI accelerates existing processes and can scale bar…
Healthcare AI has moved from pilot projects into operational use across scheduling, drug dispensing, patient communications and diagnostic decision-making, pulling nonclinical financial actors such as lenders, insurers, …
Amazon sued Perplexity AI in November 2025, alleging its Comet browser accessed Amazon.com accounts without authorization in violation of the CFAA. In March 2026, a federal judge granted a preliminary injunction against …
Anthropic kept early access to its advanced AI model Mythos for some US security firms after a US export order, while cutting off a European agency, highlighting private vendor discretion over dual-use AI governance.
The Electronic Frontier Foundation (EFF) joined Foxglove, Human Rights Watch, and 60 other organizations in urging the UK to halt its deployment of Facial Age Estimation (FAE) technology at the border, citing concerns ov…
The US government ordered Anthropic to suspend access to its Fable 5 and Mythos 5 AI models for foreign nationals due to a jailbreak vulnerability, just three days after their June 9, 2026 launch. Anthropic complied whil…
The Trump administration ordered Anthropic to cut off foreign access to its Mythos 5 and Claude Fable 5 AI models, citing national security concerns, prompting the company to take the models offline. The ban, which appli…
Meta released Pyrefly 1.0, a fast Python linter and type checker for PyTorch and Django projects. Python 3.15 introduces a sampling profiler for efficient code instrumentation. A new malware called Hades exploits Python'…
Anthropic, Microsoft, and Okta released the Enterprise-Managed Authorization (EMA) extension for the Model Context Protocol, replacing manual OAuth consent with centralized identity-driven token exchange for AI agents. T…