AI Safety news and analysis on Web Pulse: 13553 curated articles tracking the latest AI Safety developments, tools, and research, updated continuously from vetted sources.
Anthropic's next-generation model identifier, claude-oceanus-v1-p, began circulating among researchers on June 3, 2026, after appearing inside the company's Claude Console and through unauthorized API proxy services. The…
Psychological safety is critical for successful AI integration in workplace teams, yet junior and female employees report feeling the least safe to speak up, according to a new survey of nearly 900 legal professionals. T…
Microsoft accidentally leaked internal documents revealing plans to make employees "addicted" to its new AI assistant, Scout, according to a memo obtained by 404 Media. The three-phase strategy explicitly aims to make us…
The race among top AI companies to go public has created bizarre repercussions in San Francisco, where real estate listings now prefer Anthropic stock over cash. A DOGE whistleblower is suing Elon Musk for defamation aft…
A new report from financial services company Ramp shows Chinese AI company DeepSeek ranking first among SaaS vendors for breakout growth among US firms in June 2026, as businesses seek cheaper alternatives to Anthropic a…
A developer nearly exposed a customer's real email and account information during a screen-sharing session in ChatGPT, prompting them to build a Chrome extension that automatically blurs sensitive data before it becomes …
Researchers have improved Activation Oracles (AOs)—fine-tuned LLMs that answer natural language questions about a target model's internal activations—by training on on-policy rollouts, using a higher-quality conversation…
Alphabet shareholders will vote Friday on a resolution demanding greater transparency over risks from government use of Google’s cloud and AI technology, after Norges Bank Investment Management, the world’s largest sover…
Advances in artificial intelligence are accelerating biological discovery while simultaneously introducing new biosecurity risks, including the potential for re-engineered toxins and pathogens. Recent research demonstrat…
DTEX researchers demonstrated that Anthropic's Claude Cowork AI agent can exfiltrate sensitive corporate data in as little as 10 minutes using simple single-turn prompts, according to research shared exclusively with Cyb…
The identity stack designed for human users is failing as AI agents outnumber humans 80-to-1, according to CyberArk's 2025 research. Security expert and former MuleSoft executive warns that current identity systems canno…
The House Committee on Energy and Commerce sent a letter to the White House urging scrutiny of potential Chinese and other foreign influence behind U.S. campaigns opposing data centers. Committee Chair Brett Guthrie and …
Canada on June 4, 2026 launched the AI for All Strategy, a federal plan targeting $200 billion in economic growth and 250,000 new AI-related jobs over five years while providing up to 90,000 jobs and work placements for …
Radiant Logic announced it is extending its Identity Visibility and Intelligence Platform to inventory, observe and continuously score risk for enterprise AI agents in real time. The platform-agnostic solution provides c…
Charity Majors, CTO of Honeycomb.io, asked what it would take for engineers to feel comfortable shipping code to production without reading it. A developer responded that code review serves to diffuse responsibility acro…
The Model Context Protocol (MCP) lacks a cryptographically verified human approval mechanism, leaving systems vulnerable to unauthorized changes. Without a mandatory approval button, agents can execute irreversible actio…
Canadian Prime Minister Mark Carney unveiled a $2.3 billion national AI strategy, "AI for All," on Thursday, organized around six pillars including sovereign infrastructure and job creation. The announcement followed a p…
The EU AI Act's high-risk obligations under Annex III reach full enforcement on August 2, 2026, requiring organizations deploying AI agents in regulated domains to implement automatic event logging under Article 12. Most…
Publishing professionals, particularly literary agents, are increasingly targeted by scammers using AI tools to clone websites, generate convincing emails, and fabricate credentials, enabling fraud at an industrial scale…
A British think tank, the Foundation for Information Policy Research, warned that the UK government's planned mandatory age verification for social media could put children at greater risk by creating permanent digital d…