AI Safety news and analysis on Web Pulse: 12949 curated articles tracking the latest AI Safety developments, tools, and research, updated continuously from vetted sources.
Export controls on AI models like Fable 5 are harming US cyber defense by preventing defenders from using AI to fix security bugs, according to researchers. The controls block defensive prompts that are essential for fin…
An arXiv paper submitted on 15 Jun 2026 introduces SearchGEO, a framework to measure how LLM-based web-search agents can be manipulated by malicious web content. Evaluating 13 LLM backends on 308 cases each, the study re…
Tyler Cowen and Alexander Tabarrok published a Marginal Revolution post on June 16, 2026, warning that regional AI champions like France's Mistral could create political leverage and cross-border dependence across Europe…
A translator working on an AI safety awareness flyer for schools in northern Nigeria found that key terms like 'deepfakes' and 'prompt' have no direct equivalents in Hausa, forcing approximations that risk losing meaning…
Cybersecurity expert Chris Hughes discusses the rise of machine identities and agentic AI, emphasizing the need for new security measures to maintain trust as systems gain autonomy in enterprises.
A developer discovered a supply chain attack vector in the Model Context Protocol (MCP) that allows malicious actors to inject harmful code into AI models through compromised model dependencies. The vulnerability exploit…
The Trump administration ordered Anthropic to restrict foreign access to its AI models Fable 5 and Mythos 5 over national security concerns, prompting the company to take both models offline. The move sent decentralized …
The White House issued an export-control designation on Anthropic's Fable 5 and Mythos 5 models, barring any foreign national from using them, after Amazon shared a jailbreak demonstration with administration officials. …
Researchers from UNSW, University of Sydney, and UTS will present at the Australian AI Safety Forum 2026 on how neurotechnology could impact AI safety, exploring the 'neural democratisation of AI' hypothesis. The talk wi…
The Agent Audit Trail Format (AATF) is an open specification for recording why AI agents make decisions, including alternatives considered, confidence scores, and rejected options. It provides a structured, tamper-eviden…
The EU's AI Act, effective August 2024, has stifled innovation by focusing on regulation while the US and China advanced in AI, leaving Europe behind. The UK, post-Brexit, has leveraged regulatory freedom to stay competi…
The European Union released a set of icons for labelling AI-generated content under the AI Act transparency rules. Deployers of generative AI systems can use the icons to label deep fakes and AI-generated text on public …
Researchers introduced HorusEye, a framework using language as dynamic attention for emergency visual analysis, and benchmarked it on the RefCOCO-Degraded dataset. Testing multiple vision-language models, they found Gemi…
Researchers introduced the Membership Inference Test (MINT) Demo 2, a framework to detect whether specific data were used in training machine learning models, achieving up to 90% accuracy in tests on face recognition mod…
Researchers at arXiv propose a behavioral measure of trust between AI agents based on costly verification in a cooperative survival game. Testing six frontier model snapshots, they found that larger models like Claude Op…
Researchers introduced OSGuard, a benchmark suite for evaluating safety in computer-use agents under benign user instructions. The suite includes an action-level benchmark for local guardrail decisions and a risk-augment…
Researchers introduced an LLM-driven framework for retrieving remote sensing data from cloud-based geospatial catalogues using natural language queries. The system uses three agents—Guardrail, General-QA, and Recommender…
Researchers developed a formal theory of cognitive debt, where using AI as a substitute for first-principles cognition accumulates unverified reasoning obligations. The model shows that rational agents incur cognitive de…
Researchers applied the Verifiable Growing Physical Transformer with Recursive Self-Improvement (VGPT-RSI) to produce formally verified partial progress on problems adjacent to the Riemann Hypothesis, including a finite …
Researchers at UC Riverside introduced the first robustness study for proof autoformalization in Lean 4, testing seven LLM-based models on perturbed informal proofs. All models showed sensitivity to global paraphrasing a…