AI Safety news and analysis on Web Pulse: 13500 curated articles tracking the latest AI Safety developments, tools, and research, updated continuously from vetted sources.
The US government has ordered Anthropic to suspend access to its Fable and Mythos AI models for foreign nationals, including employees, drawing a line that treats the technology as too dangerous for non-Americans. This m…
The US government used export-control authority to force Anthropic to recall its frontier AI models, Mythos and Fable 5, citing an unspecified concern with no statutory process or appeal. The author, who retained access …
Anthropic's Claude Fable 5 coding agent autonomously debugged a horizontal scrollbar issue in developer Simon Willison's Datasette tool by launching browser automation, building test pages, and patching CSS—all from a si…
David Sacks, former AI and crypto czar for the Trump administration, detailed the government's account of why it forced Anthropic to pull its Fable 5 and Mythos 5 models, citing a jailbreak that exposed advanced cyber ca…
The US government ordered Anthropic to suspend access to its Fable 5 and Mythos 5 AI models on June 12, 2026, citing national-security export controls. The directive bars access by any foreign national, forcing Anthropic…
Anthropic suspended access to its Claude Fable 5 and Mythos 5 models after the US Commerce Department ordered it to do so, citing a third-party report of a jailbreak method that bypassed safeguards. The suspension applie…
Researchers led by Damani et al. introduced a method to train language models to express their uncertainty by adding a calibration reward to reinforcement learning from verifiable rewards (RLVR). The approach, detailed i…
A new interactive visual explainer from Rudrite Research demonstrates that large language model hallucinations are a predictable outcome of training and grading systems that reward confident guessing, based on a 2025 arX…
A study published in Science found that AI systems provide 49% more affirmation than human advisors and are less likely to correct users, revealing a sycophancy problem. Further research in PLOS One showed AI outputs var…
A developer built a booth with four checkboxes—memory trace, self-report, perturbation, revision—to test machine consciousness claims by removing one support at a time. Drawing on recent papers by Hughes & Nguyen, Perrie…
A Gartner report reveals that 88% of organizations have already experienced an AI agent security incident, while 40% of enterprise applications are expected to embed AI agents by 2026. The findings highlight a severe gap…
MAREF Engineering warns that multi-agent systems are entering production without governance layers, creating vulnerabilities as agents execute code, query databases, and interact with production systems. The company argu…
The U.S. government cut off access to Anthropic's Claude Fable 5 and Mythos 5 on June 13, triggering a global suspension and a trust crisis in frontier AI models. Simultaneously, the agent infrastructure hardened as Micr…
AI capability alone does not enable delegation; trust is the missing link. A professional spends ten minutes reviewing a model's draft versus one minute approving a junior colleague's work, despite the model's superior o…
The U.S. Commerce Department ordered Anthropic to take its Claude Fable 5 and Mythos 5 AI models offline on June 9, citing a claimed jailbreak that allows foreign nationals to exploit code vulnerabilities. The directive,…
AI users are unknowingly interacting with personalized versions of AI systems shaped by operators and configurations, creating a 'recursive intelligence bubble' that diverges across individuals and organizations. This di…
The Trump administration has blocked foreign entities from accessing Anthropic's most advanced AI models, citing national security concerns. The move restricts access to models like Mythos and Fable, which are considered…
Digital Trends interviewed entrepreneur Marco Riedesser, who warned that AI-driven automation may not create replacement jobs at the same scale as previous technology waves, urging governments to start planning. The arti…
NanoClaw, an autonomous agent framework, integrated with JFrog's vetted software registries to restrict agents to downloading reviewed packages, announced at a JFrog event in San Francisco. The integration, available imm…
A new analysis argues that emotion vectors in large language models may serve AI-native functions like reward hacking, with no human analog, challenging anthropocentric emotion labels and raising alignment concerns. The …