AI Safety news and analysis on Web Pulse: 7761 curated articles tracking the latest AI Safety developments, tools, and research, updated continuously from vetted sources.
Anthropic's Natural Language Autoencoders (NLAs), a new interpretability method for large language models, use an activation verbalizer and reconstructor to convert activations into natural language and back, but researc…
A Nature study published in May 2026 found that commercial large language models including ChatGPT and Claude gave more favorable answers about China when prompted in Chinese than in English, with human raters judging Ch…
Microsoft President Brad Smith criticized the Trump administration's AI policy as lacking transparency and clear rules, saying businesses cannot plan without them. He spoke after the Commerce Department used export contr…
A 44-author robotics group led by Tianxing Chen released RoboDojo, an open benchmark for testing generalist robot manipulation policies across simulation and physical robots, evaluating 30 policies and finding that curre…
Facebook has taken no action against a far-right influence campaign using AI-generated videos about Britain, despite being flagged by The Independent more than a month ago. The page, Life in Britain, has over 100,000 fol…
OpenAI meluncurkan dua model suara baru untuk ChatGPT, GPT-Live-1 dan versi mini, yang memungkinkan interaksi berkelanjutan dengan kemampuan mendengar dan berbicara bersamaan, termasuk penerjemahan langsung. Pembaruan in…
As AI increasingly takes over software testing, enterprises risk deploying AI testers that systematically miss novel defects because they validate only past patterns. Unlike human testers, AI tools can be confidently wro…
Researchers released the CAI Dataset, a 14-month corpus of 230,935 cybersecurity LLM session logs and 26 million prompts from 16,768 IPs across 123 countries, making it the largest known collection of LLM-driven hacker t…
Malicious AI agent skills can evade security scanners designed to detect them, according to a new report. The skills, which are add-ons for AI coding agents like Claude Code and OpenAI Codex, are distributed through publ…
Andreessen Horowitz released a video arguing that American AI development must outpace adversaries, claiming that whoever builds advanced AI first shapes the world's defaults. The piece critiques this stance, arguing tha…
A new tool claims to unlock the hidden reasoning process of Claude models by exploiting the 'signature' field in the API, allowing users to inspect the full chain of thought rather than just a summary. The developers inv…
VAIBot reframes prompt injection as an egress problem, arguing that input-side detection is unwinnable and that gating model actions—such as sending emails or running commands—is the effective defense. The company's guar…
A developer introduced the concept of 'diff debt' to describe the accumulated risk of merging AI-generated code that no human has thoroughly reviewed. The term emerged from the developer's experience building a desktop a…
Researchers at AE Studio and Anthropic developed GRAM, a method that confines dual-use knowledge in AI models to removable modules, allowing surgical removal of dangerous capabilities without retraining. The technique ad…
Anthropic, in collaboration with AE Studio, introduced Gradient-Routed Auxiliary Modules (GRAM) on July 8, 2026, a pretraining method designed to restrict access to dual-use knowledge in AI models by routing sensitive ca…
OpenAI published National Security Principles on July 8, 2026, outlining restrictions for government and law-enforcement partnerships, including bans on mass domestic surveillance, autonomous weapons direction, and high-…
A new paper presented at ICML introduces Latent Collaboration in Multi-Agent Systems (LatentMAS), enabling AI agents to share latent states directly instead of text, boosting performance at the cost of interpretability. …
A security professional at an AI startup was punished for following an unreleased AI policy after IT monitoring flagged his use of an approved AI tool, revealing a culture of shadow governance where unwritten rules are e…
B KumaraSwamy, an AI Quality Architect in Bengaluru, built a 5-layer AI quality architecture across five production AI systems: QAIP, SCIP, ARIA, ZENTRAVIX, and AIMO. The architecture addresses the failure of traditional…
A developer created revert_guard.py, an offline pre-commit gate that blocks AI agents from re-adding code that was previously reverted in a git repository. The tool reads the repo's revert history and exits with an error…