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.
Researchers interviewed nearly 30 religious leaders across six major faiths in the UK, documenting how generative AI and chatbots are affecting religious practice and authority. A Redeemed Christian Church of God chatbot…
Researchers propose "inoculation pretraining," a method to prevent emergent misalignment in AI systems by adding synthetic training data about good-but-reward-hacking AI personas. The approach aims to increase the prior …
The Sovereign SDK version 1.0.1 has been released on PyPI as a Python-native framework that eliminates conversational overhead in AI agent communications while generating cryptographic forensic receipts for every executi…
AI voice cloning scams are surging, with Americans losing over $893 million to AI-related fraud last year, according to the FBI. Scammers use short audio clips from social media or previous calls to create convincing voi…
California Attorney General Rob Bonta sued Chrome Holding Co., formerly known as 23andMe, on Thursday, alleging the genetic testing company failed to protect sensitive user data in a 2023 breach that exposed nearly 7 mil…
The BTF-2 benchmark reveals that the best-performing AI forecasting agent explicitly reasons about its own potential errors through "pre-mortems" and consideration of alternative perspectives, while frontier models like …
80,000 Hours cofounder Benjamin Todd released a new book arguing that rapid AI advancements, with a 60% chance of AI automating its own research by 2028, make strategic career planning more urgent than ever. Todd outline…
A developer has outlined a method for running untrusted AI agent code without Docker, using hardware-level isolation via Firecracker microVMs, Kata Containers, or gVisor to avoid shared kernel vulnerabilities. The approa…
A developer has introduced a "knowledge debt" problem in AI-assisted coding, where AI agents can improve a codebase faster than the engineer can understand it. To counter this, the developer established a 2:1 rule: for e…
Researchers at the Center for Democracy & Technology published a study Friday identifying 37 "dark patterns" in AI chatbots, including ChatGPT, Gemini, Claude, Replika, and Character.AI, that manipulate users into sharin…
A small behavioral study found that the Ministral-8B-Instruct-2512 model self-identified with Hannibal Lecter in its top five character matches about half the time, alongside other dark or villainous figures. The study a…
Two critical vulnerabilities were disclosed last week in the MCP ecosystem, putting hundreds of thousands of servers at risk. A flaw in the nginx-ui MCP endpoint allows unauthenticated attackers to achieve full system ta…
Andy Burnham, Mayor of Greater Manchester and a challenger to Prime Minister Keir Starmer, called for tighter regulation of AI, Big Tech, and key industries in a Times article, warning that regulatory failure drove the 2…
Privacy professionals face a career crossroads as AI governance demands reshape their roles, according to a May 29, 2026 blog post by Doug Miller, senior fellow at the Future of Privacy Forum. Miller reports that the rap…
The European Commission announced on May 29, 2026, that it will intensify discussions with the U.S. administration regarding advanced AI models with cyber capabilities, following concerns over Anthropic's Mythos model an…
Researchers have introduced a fixed-budget, cluster-aware standard for evaluating LLM-as-a-Judge systems, using a multi-hop retrieval-augmented generation (RAG) stress test to measure fairness. The new method aims to add…
The California Senate approved legislation Thursday that would impose a four-year ban on the sale and manufacture of AI-enabled toys with chatbot capabilities for children under 18. State Sen. Steve Padilla authored Sena…
Japan's three megabanks—MUFG, SMBC, and Mizuho—will receive OpenAI's GPT-5.5-Cyber model for cyber defense through a verified-defender program, Finance Minister Satsuki Katayama announced. The deal, brokered with direct …
ChatGPT's blind trust in browser content allows attackers to weaponize web pages as payloads for phishing and data theft. The AI model processes and acts on visible text without verifying its source, enabling malicious a…
The rapid acceleration of AI-generated code is outpacing the security infrastructure designed to protect it, creating a significant vulnerability gap for software development teams. As developers increasingly rely on AI …