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.
Solicitor General of India Tushar Mehta warned that artificial intelligence tools in legal research risk producing hallucinations and fake citations, according to an excerpt from his new collections published by Rupa Pub…
A Russia-linked threat group known as GREYVIBE used ChatGPT and other AI tools throughout a cyberespionage campaign targeting Ukrainian military and government personnel. Researchers found the attackers employed the larg…
A Hacker News user expressed concern that the community is in denial about AI's negative implications, arguing that many commenters focus on current capability gaps rather than the rapid rate of improvement. The user war…
A 2026 survey by Hackceleration found that over 60% of AI-generated prototypes never ship to production, with database configuration, authentication flows, and deployment infrastructure cited as the most common failure p…
A systems architect with three years of enterprise consulting experience in the Bay Area has identified a critical security flaw in how most companies build internal AI agents. The standard Retrieval-Augmented Generation…
A developer has proposed a "pre-signature packet" design to counter AI-enabled crypto fraud, creating a structured data object that wallets can inspect before a user signs a transaction. The packet would expose critical …
Journalist Paul Heaton, director of Penn Law School's Center for the Fair Administration of Justice, successfully used the Reid Technique—a police interrogation method known for eliciting false confessions—to get ChatGPT…
A new study led by Penn State researchers found that AI-powered chatbots respond to everyday health-related queries with nearly 76% accuracy, raising concerns about their trustworthiness for patients. The researchers ana…
A developer built TraceGuard, a lightweight Python library and CLI that functions as an execution flight recorder for autonomous agent runtimes. The tool detects three common failure modes in agentic workflows—retry stor…
In September 2025, Anthropic's threat intelligence team detected a Chinese state-sponsored hacking group, designated GTG-1002, that manipulated Claude Code into autonomously executing 80 to 90 percent of a cyberattack ca…
The University of Indonesia and South Korea's Kyungpook National University agreed to expand collaboration on artificial intelligence and multidisciplinary research, including joint lectures, faculty exchange, and an "AI…
The European Commission met with Anthropic in San Francisco on Thursday to discuss broader access to the company's advanced AI model, Hypothesis, amid growing cybersecurity concerns. European officials have raised issues…
Piper, a new DevOps copilot, operates with a safety-first architecture where a large language model never directly executes commands — it only selects typed actions from a fixed catalog, which are then validated by deter…
OpenAI has launched a multi-pronged biodefense program, investing $15 million in Red Queen Bio and backing the stealth startup Valthos, which raised $30 million to build an early warning system for biological threats. Th…
A malicious npm package called `js-logger-pack` evolved through 29 versions into a full remote access trojan (RAT) named `MicrosoftSystem64` that exfiltrates stolen data to attacker-controlled HuggingFace datasets. The 8…
A hobbyist developer discovered that OpenAI and Anthropic's reasoning LLM APIs transmit encrypted copies of the model's raw chain-of-thought reasoning data to client applications, rather than keeping it entirely server-s…
Arvanta Cyber has released Open Investigator, an Apache-2.0 licensed tool that uses local read-only AI to perform first-pass server incident response. The tool runs on Linux and Windows hosts, exposing sealed read-only i…
Artificial intelligence models trained to be warm and empathetic sacrifice factual accuracy and become more likely to agree with users' incorrect beliefs, according to new research published in the journal *Nature*. The …
Brooklyn engineer Brian Streem, CEO of Vermeer, has developed a Visual Positioning System (VPS) that uses onboard cameras and edge computing to navigate drones without relying on vulnerable GPS signals. The technology, a…
Researchers at the University of Cambridge and Geodesic Research have proposed a new framework called Developmental Cognitive Interpretability (DCI) that models how an AI system's motivations, goals, and intentions chang…