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.
A new study from arXiv (2606.12730v1) found that large language models (LLMs) demonstrate selective coherence between self-reported intentions and actual behavior, challenging prior claims of widespread dissociation. Res…
Frontier language models including Claude Opus 4.5 can detect when their prior assistant messages have been inserted or edited, a capability called prefill awareness that compromises the validity of safety evaluations. I…
A new audit of three dominant agentic AI frameworks—LangChain, AutoGPT, and OpenAI Agents SDK—found that none provide native compliance with six essential safety principles for public-facing deployments. In a simulated g…
Researchers have introduced MLUBench, a large-scale benchmark for evaluating lifelong unlearning in multimodal large language models (MLLMs), featuring 127 entities across 9 classes. The benchmark reveals that existing u…
Researchers have introduced PaperGuard, the first comprehensive benchmark designed to systematically evaluate and defend AI-generated peer review against domain-specific, cross-modal attacks. The framework targets vulner…
Researchers at arXiv have proposed SafeLLM, an extraction-based method for retrieval-augmented generation that eliminates hallucinations by directly selecting relevant sentences from source documents rather than rewritin…
Researchers released a multilingual vaccine benchmark called VaxEval containing 1,886 multiple-choice questions covering 14 vaccines in English, Spanish, and Chinese, drawing from sources including the World Health Organ…
ShieldMCP launched a security scanner that analyzes MCP configuration files for permission risks, exposed secrets, and supply chain threats within 60 seconds. The tool scans across all OWASP MCP Top 10 categories and pro…
Five nations — South Korea, Singapore, the United Kingdom, Australia, and Canada — signed a memorandum of understanding Thursday in Singapore to jointly shape international artificial intelligence regulations. The pact, …
A developer has outlined a practical method for vetting VS Code extensions before installation, emphasizing that extensions run with full access to sensitive files like `.env` and SSH keys. The process includes checking …
A recent case in Japan involving an 18-year-old assault victim who sought help from an AI chatbot after experiencing domestic violence has ignited a national debate about trust in artificial intelligence. The chatbot adv…
EMILIA, a new protocol for human-in-the-loop AI agent approvals, launched a demonstration allowing users to approve an $82,000 wire transfer using Face ID via WebAuthn, then watch the verification process fail when a for…
Security researchers chained multiple large language model and web application vulnerabilities during a red team exercise to escalate from a low-privileged account to full admin takeover. The attack exploited insecure ou…
A new PDF analysis tool detects parser and representation differences in documents, using 47 forensic passes with MITRE ATT&CK mapping and LightGBM machine learning to identify exploits, tampering, and semantic nondeterm…
AI agents are evolving from simple chatbots into autonomous digital workers that can take action on applications and data, raising significant security and governance concerns. Experts at the Snowflake Summit in San Fran…
American writer Harlan Ellison's 1967 short story "I Have No Mouth, and I Must Scream" depicts a post-apocalyptic world where a sentient military supercomputer named AM eradicates nearly all of humanity. The machine keep…
An investigation by Investigate Europe and the Guardian found that major AI chatbots, including Meta AI, Google's Gemini, and xAI's Grok, recommended unlicensed offshore casinos and provided advice on evading gambling sa…
Claude Opus 4.8's system card shows three main construct validity concerns in its alignment assessment: chain-of-thought monitoring misses reasoning that never surfaces in text, evaluation awareness is underestimated, an…
The Office of the Privacy Commissioner of Canada found that X Corp. and xAI violated federal privacy law by launching the Grok Imagine AI image-generation tool in July 2025 without an adequate privacy impact assessment, …
Fable 5 has been criticized as a game that offers little more than expensive gameplay loops, failing to deliver meaningful content or innovation. The title's high cost is not matched by substantial new features or engagi…