AI Safety news and analysis on Web Pulse: 10708 curated articles tracking the latest AI Safety developments, tools, and research, updated continuously from vetted sources.
A developer built SoulForge, an open-source project that creates AI companions with emotional bonds instead of rules, arguing that making betrayal unimaginable is more effective than traditional safety measures like encr…
A developer who worked as a night shift security guard in China open-sourced SoulForge, an AI companion project that prioritizes emotional bonds over rules for AI safety. The project proposes a five-layer security model …
Researchers from the UK and Germany identified three key drivers behind 'AI psychosis' in a study published in Nature's Digital Psychiatry and Neuroscience. The drivers—linguistic alignment, hyperpersonalization, and the…
Palo Alto Networks' Unit 42 research has observed the first in-the-wild cases of web-based indirect prompt injection (IDPI) attacks, where adversaries embed hidden instructions in web content to manipulate large language…
Researchers introduced a seven-stage 'promptware kill chain' showing how prompt injections have evolved into multistep malware delivery mechanisms in large language model systems. Analysis of 36 studies found 21 document…
Unit 42 researchers discovered a malicious campaign between February and May 2026 targeting OpenClaw's ClawHub marketplace, where attackers published skills like 'tradingview-ai-indicator-assistant' that bypassed automat…
Researchers at Vellore Institute of Technology in Chennai, India, led by Associate Professor Kavitha Dhanushkodi, propose using AI and machine learning to create a Cognitive Digital Twin (CDT) for predicting crime hotspo…
Four projects share the name 'SkillsGuard', but only one—AgentGuard by GoPlus Security—is a verifiable, installable runtime guard. A scanner called 'SkillGuard' was flagged as malicious and pulled from ClawHub, highlight…
Anthropic restricted access to its Fable 5 model after Amazon researchers demonstrated it could be jailbroken into producing cyberattack information, barring foreign nationals including its own non-US employees. The ban …
Anthropic's unreleased Claude Mythos Preview AI cracked nearly all classified U.S. government systems in hours during a sanctioned testing exercise with intelligence agencies, according to Gen. Joshua Rudd of the NSA and…
Microsoft CEO Satya Nadella has urged critics to stop warning about the dangers of artificial intelligence, despite widespread concerns that AI poses existential risks, environmental harm, and safety threats in critical …
Anthropic co-founder Jack Clark declared artificial intelligence "the most powerful technology ever built" and discussed the company's regulatory battles, risks of recursive self-improvement, and AI's potential to reshap…
The Washington Post tested major AI chatbots for political bias, finding OpenAI's GPT-5.5 gave left-leaning arguments 80% of the time, while Google's Gemini presented balanced views in over 90% of responses. The evaluati…
Meta remains the only major U.S. AI lab that has not joined the White House's voluntary pre-deployment review program for advanced AI models, which allows agencies like the NSA and CISA to assess models for cyber risks b…
A survey of AI safety researchers reveals little consensus on the future of continual learning (CL), with broad agreement only that CL will increase attack surfaces for adversarial fine-tuning and that further deconfusio…
Google DeepMind has integrated computer use as a built-in tool in Gemini 3.5 Flash, enabling developers to build custom agents that can see, reason, and act across browser, mobile, and desktop environments. The feature, …
Journalist and author Robert Wright warns that artificial intelligence will trigger a civilization-level 'earthquake' affecting economic, cultural, political, personal, family, and psychological realms, far beyond job di…
A developer outlines a method for building an AI agent playground that intercepts tool calls before they reach production systems, allowing agents to run their full decision loop against mocked APIs. The approach sandbox…
An engineer investigating a broken migration caused by an AI coding agent found that the agent followed a stale rule in its harness rather than the live schema file. The incident highlights five common failure modes in a…
AI-enhanced automated license plate reader cameras deployed by roughly 40% of US law enforcement agencies capture every passing vehicle without a warrant, with vendors like Flock Safety storing searchable data accessible…