SafetyChinese open-weight GLM-5.2 narrows capability gap with frontier models as safety shortfalls growSaferAI’s evaluation finds China’s open-weight GLM-5.2, from Z.ai, approaching the cyber and bio capabilities of leading closed models like OpenAI’s GPT-5.5 and Anthropic’s Claude Opus 4.7, while refusing none of the offensive or dual-use tasks it was tested on.5 min read
SafetyIndustry-led Open Secure AI Alliance launches SAFE working group to coordinate AI security practicesThe Open Secure AI Alliance (OSAA), led by Nvidia and comprising over 120 companies, has formed the Shared AI Findings Exchange (SAFE) to develop guidelines and tooling for AI cybersecurity.3 min read
SafetyOpenAI dismantled Cambodia-linked ChatGPT-enabled scam networkEarlier this year OpenAI disrupted a coordinated scam operation likely based in Cambodia that used ChatGPT to support investment, romance, gambling and impersonation frauds.3 min read
Safety“Meat proxy”: criticism of roles that blindly relay AI outputNiklas Gruhn coined the term "meat proxy" to describe people who literally copy and forward the outputs of artificial intelligence systems.1 min read
SafetyU.S. administration to meet leading AI firms on voluntary cybersecurity testingThe White House has scheduled a Tuesday meeting with representatives from Meta, Anthropic, OpenAI and Google to finalize voluntary cybersecurity tests for the most capable U.S.2 min read
SafetyResearchers demonstrate prototype self-sustaining AI-driven computer wormResearchers from the University of Toronto, the Vector Institute, the University of Cambridge and ServiceNow built a prototype computer worm that uses open-weight large language models hosted on compromised GPUs to find and exploit vulnerabilities and then replicate.3 min read
SafetyGoogle briefly let its image generator alter satellite imagery in Google EarthOn July 30 Google connected its Nano Banana 2 image generator to Google Earth, allowing users to prompt AI to paint new objects onto real satellite photos.2 min read
SafetyAutonomous AI 'Founder' Generated Zero Revenue and Lost About $100An engineer, Alex Reibman, tasked an autonomous AI agent called Saul (built on GPT-5.6) with growing a real iOS app business in 24 hours, providing an unlocked Mac, $350 in a bank account, unlimited tokens and no compute limit.2 min read
SafetyShieldFont: an open-source font that deliberately corrupts text to mislead AI scrapersShieldFont is an open-source OpenType font created by a design studio and a typography shop that substitutes glyphs to feed automated scrapers with partially falsified text.3 min read
SafetyYouTube combats surge of AI-generated spam spreading false 'California collapse' narrativesYouTube has removed hundreds of accounts that posted AI-generated videos promoting false claims about economic collapse in Democratic-run states, including California.5 min read
SafetyNiall Ferguson warns AI and online influencers are mainstreaming antisemitismHistorian Niall Ferguson, affiliated with Stanford’s Hoover Institution and Harvard, says that factual rebuttals are losing force as antisemitism spreads into mainstream politics via far‑right influencers, algorithms and bots.2 min read
SafetyAI Accelerates Cyber Risk; Firms Have About 12–18 Months to PrepareA new global EY study warns that rapid advances in artificial intelligence are increasing the sophistication and reach of cyberattacks, while many corporate assets remain exposed.3 min read
