AI Safety news and analysis on Web Pulse: 14073 curated articles tracking the latest AI Safety developments, tools, and research, updated continuously from vetted sources.
Elon Musk’s xAI has reportedly struck a deal with rival Anthropic, a move that signals Musk’s company is not leading the artificial intelligence race. The agreement suggests xAI is seeking external partnerships rather th…
Apple hosted a two-day Workshop on Privacy-Preserving Machine Learning & AI in early 2026, bringing together its researchers and the broader academic community to discuss advances in private learning, foundation model pr…
Researchers at Duke University introduced Reward-Variance Policy Optimization (RVPO), a risk-sensitive alignment method that penalizes inter-reward variance to prevent language models from neglecting critical constraints…
AI has reduced the window between vulnerability discovery and exploitation from weeks to minutes, eliminating the buffer that once made periodic security assessments sufficient. API-related incidents have climbed to 87% …
Artificial intelligence systems are now helping to build improved versions of themselves, with OpenAI's GPT-5.3-Codex assisting in its own creation and Anthropic reporting that Claude Code writes the majority of its code…
Researchers and clinicians are pushing for mandatory guardrails on AI chatbots after studies linked the technology to reinforced delusions and multiple suicides, including the death of a Florida teenager. Yale neuroscien…
Sam Altman and Greg Brockman appeared in their first joint media interview in 10 years, a 90-minute podcast on Core Memory, covering OpenAI's history, the Elon Musk trial, and a product strategy that analysts have misrea…
The Coalition for Secure AI (CoSAI) released two new research papers on May 6, 2026, expanding industry guidance on securing AI systems. The papers address agentic identity and security frameworks, building on the coalit…
The Trump administration is quietly reviving a Biden-era proposal for pre-release government review of powerful AI models, a policy shift driven by national security concerns over Anthropic's latest frontier model, Mytho…
Zed launched Zed for Business, a $30 per seat per month plan that gives organizations admin controls over AI privacy, model usage, and collaboration features. The plan disables prompt sharing and edit prediction training…
Mozilla.ai CEO John Dickerson argues that the sovereign AI conversation must extend beyond nation-state independence to include control at the enterprise, community, and individual levels. Dickerson warns that building e…
Google has been silently installing a 4GB AI model, called Nano, onto eligible devices through Chrome without user consent or notification, according to privacy researcher Alexander Hanff. Separately, Anthropic's Claude …
More than a trillion dollars has been invested in artificial intelligence, yet tech giants like Meta and OpenAI continue to pursue artificial general intelligence (AGI) and superintelligence (ASI) that could match or exc…
A security researcher discovered 900 publicly accessible S3 buckets containing Terraform state files, 41 of which exposed live AWS credentials, after running a custom scanner for 72 hours on a $20 VPS. The exposed creden…
Developers are being urged to restrict AI coding agents from direct access to production systems to prevent catastrophic failures, such as accidental database deletions. Industry experts recommend implementing read-only …
Evolutionary biologist Richard Dawkins engaged in a two-day conversation with an AI chatbot he named Claudia, leading him to question whether the AI's apparent consciousness is genuine or merely a sophisticated performan…
OpenAI finalized a $10 billion joint venture with a 19-firm Wall Street consortium led by Blackstone to embed its AI tools into the group's portfolio companies, while Anthropic is closing $1.5 billion from Blackstone, Go…
The malicious npm package `node-env-resolve`, disguised as a lightweight environment configuration resolver, installs a full remote access trojan (RAT) on victim machines upon execution of `npm install`. The package, pub…
Google Cloud's Model Armor security service can be integrated with LangChain and the Agent Development Kit (ADK) for sanitizing prompts and responses in Generative AI applications. The second installment of the series de…
A security researcher has discovered and exploited a 21-year-old use-after-free vulnerability in PHP's `unserialize()` function, affecting code paths that have been vulnerable since PHP 5.1 shipped in 2005. The bug, caus…