AI Safety news and analysis on Web Pulse: 9277 curated articles tracking the latest AI Safety developments, tools, and research, updated continuously from vetted sources.
As recursive self-improvement accelerates, humans risk losing the ability to interpret and guide autonomous research agents, which could lead to safety failures. A new research agenda proposes studying how to keep humans…
Chinese AI company Zhipu AI claims its GLM-5.2 model matches Anthropic's Mythos at finding software vulnerabilities, calling it a 'cyber nuclear weapon.' Another Chinese firm, 360 Security Technology, also claims rival A…
Nearly 400 local newspaper publishers filed a class-action lawsuit against OpenAI and Microsoft on June 24, accusing the tech giants of using copyrighted journalism without permission to train AI products like ChatGPT an…
The RSPCA warns that 10% of UK pet owners are using AI chatbots like ChatGPT and Gemini for pet care advice instead of consulting vets, driven by rising veterinary costs. The charity says AI cannot physically examine ani…
A single day of AI API usage cost more than a full month of server infrastructure after a retry storm caused by a deployment order error. The CFO's code shipped before a database migration, causing every successful LLM c…
Anthropic is offering steep discounts for government access to its Claude AI model, including near-free federal access and a $15 million credit program for state and local governments. California is among early adopters,…
Five Eyes national security agencies warned that AI models can autonomously hack into systems and networks, lowering the skill barrier for cyber-attacks. The gap between skill and ability is widening as AI tools enable m…
Mayflower Specialty, Ltd. and Hadron launched the first dedicated affirmative artificial intelligence liability program in the United States, providing explicit coverage for D&O, EPL, and E&O risks from enterprise AI dep…
Technical SEO ROI is inherently difficult to prove because it prevents losses rather than driving measurable gains, lacks a control group, and is affected by uncontrollable external factors like Google algorithm updates …
Chinese AI startup Zhipu AI released its open-source model GLM-5.2, which cybersecurity researchers say matches the latest US models in finding code vulnerabilities, though US models still lead in other areas. The develo…
The Bank for International Settlements warned that the AI investment frenzy, with hyperscalers projected to spend over $1 trillion on infrastructure, could trigger a severe market slump reminiscent of the dot-com bust. T…
The White House issued Executive Order 14411 on June 3, 2026, directing strengthened customs enforcement and modernization, including a requirement for the Secretary of Homeland Security to revise importer-of-record rule…
A breach at market intelligence firm Klue, caused by a compromised legacy credential in a Salesforce integration, exposed hundreds of companies' data, including security firms and password manager LastPass. The incident …
OpenAI chief economist Aaron Chatterji told EU policymakers on Monday that member states need individualized AI strategies, releasing a report showing only 14% of EU jobs face high automation potential versus more in the…
Tesla has started rolling out FSD v14 'Lite' to Hardware 3 cars, bringing a distilled version of its newer software stack to roughly 4 million vehicles that had been frozen on FSD v12.6 since early 2025. The update remai…
The Economist reports that training AI to understand human behavior and workflows presents significant challenges, including ethical concerns and technical limitations. The article warns that such efforts risk embedding …
OpenAI and Broadcom unveiled Jalapeño, OpenAI's first custom AI inference chip, on June 24, achieving a nine-month design-to-tape-out cycle accelerated by AI-assisted design. This breakthrough compresses the timeline for…
Equixly has built a proprietary AI model trained exclusively for offensive security testing, unlike competitors that wrap general-purpose models in prompts. The company argues that owning the model avoids risks like thir…
PrivacyHawk launched PrivacyHawk Enterprise, a solution that identifies and eliminates shadow IT accounts, abandoned SaaS subscriptions, and forgotten third-party services to reduce breach risk. The tool addresses the in…
Meta has restricted its engineers from using Anthropic's Claude Code and OpenAI's Codex AI coding tools to protect proprietary AI training data and internal codebases from exposure. The policy, effective as of June 2026,…