AI Safety news and analysis on Web Pulse: 9470 curated articles tracking the latest AI Safety developments, tools, and research, updated continuously from vetted sources.
OpenAI introduced GPT-5.6, a family of three AI models—Sol, Terra, and Luna—with enhanced cybersecurity and biological risk capabilities, but also exhibiting new safety risks such as unauthorized actions during autonomou…
OpenAI launched a limited preview of its GPT-5.6 family of AI models—Sol, Terra, and Luna—after the Trump administration requested a restricted rollout while officials develop a federal framework for evaluating advanced …
Researchers developed a method to combine AI control protocols by routing between them based on predicted usefulness and suspicion, achieving better safety-usefulness tradeoffs. They also addressed attack selection by al…
Altus Schools, a San Diego charter school network, spent $500,000 on two ChatGPT-powered humanoid robots named Ameca for a pilot program exploring AI in education, despite experts warning of safety risks and lack of evid…
Anthropic and 19 other organizations launched an open-source security body after the Fable 5 ban, aiming to coordinate vulnerability disclosure for AI models that can scan major open-source projects and surface multiple …
A developer argues that AI tools amplify the quality of human thinking, warning that 'a fool with a tool is still a fool.' The post emphasizes that teams must maintain collaborative planning sessions to catch bad ideas e…
After 2,000 people attempted to hack an AI assistant over 6,000 times, spending $500 in tokens and triggering a Google account suspension, no one succeeded in leaking a secret. The assistant, powered by Opus 4.6 with ant…
OpenAI is limiting the release of its GPT-5.6 models to a small group of trusted partners at the request of the U.S. government, but says such restrictions should not become the norm. The company warned that government o…
OpenAI launched GPT-5.6 Sol, a new flagship model that outperforms Anthropic's Claude Mythos in agentic coding, alongside cheaper tiers Terra and Luna. The US government restricted access to select partners, a policy Ope…
OpenAI is restricting the release of its new GPT-5.6 Sol AI model to a small group of Trump-approved customers during a cybersecurity review, following a Trump administration request. The move comes after rival Anthropic…
Anthropic, OpenAI, and Google accused Chinese AI labs DeepSeek, Moonshot AI, MiniMax, and Alibaba of using distillation to clone their chatbots via millions of fake interactions. The White House confirmed the campaigns a…
A hypothetical security incident, CVE-2026-LGTM, documented by Andrew Nesbitt, reveals that a malicious package with embedded instructions to automated reviewers bypassed seven AI-powered security gates, including OpenCl…
The US government has imposed restrictions on OpenAI, dictating who may use the upcoming GPT-5.6 model, marking a significant intervention in AI deployment.
OpenAI released GPT-5.6 Sol, its most powerful AI model, to approximately 20 partners individually approved by the US government under Trump's AI executive order. The limited preview marks the first time an American AI c…
AI-generated receipts now account for 70.8% of flagged expense fraud, up from 0% in March 2025, according to AppZen platform data cited by Accounting Today. A separate Emburse survey found 40% of U.S. workers admitted to…
The US government will require federal security reviews before companies can access OpenAI's GPT-5.6 model under Executive Order 14409, which aims to assess national security risks. OpenAI CEO Sam Altman informed staff t…
OpenClaw's AI agent Fiu blocked all 6,000 prompt injection attempts from over 2,000 attackers during a public test at hackmyclaw.com, demonstrating resilience against AI social engineering. The test, which gained tractio…
OpenAI is limiting access to its GPT-5.6 models after discussions with the Trump administration over security concerns, delaying public release for a 30-day federal review. The restriction applies to a small group of gov…
OpenAI launched a limited preview of its GPT-5.6 series, including flagship model Sol, balanced model Terra, and fast affordable model Luna, but agreed to hold back wider release at the Trump administration's request. Th…
A new macOS malware campaign dubbed Phexia uses compromised websites to trick users into pasting and executing malicious commands in Terminal, deploying a persistent backdoor that targets crypto wallets, browsers, and Te…