AI Safety news and analysis on Web Pulse: 9365 curated articles tracking the latest AI Safety developments, tools, and research, updated continuously from vetted sources.
Cate Blanchett launched the Human Consent Registry at the European Parliament, a free website that allows individuals to protect their likeness from unauthorized AI use. The tool, presented with MEP Eva Maydell and direc…
Solo developer launches AgentWatch, a tool that sits in front of major AI model providers to enforce runtime budgets and prevent runaway loops and unexpected costs in AI agents. The open-source project aims to solve a co…
The Linux kernel in Q2 2026 removed old hardware drivers including i486 CPU support and eliminated the strncpy API after six years of work, while AI/LLM-detected security vulnerabilities drove some changes. Performance o…
Researchers at Irregular, a frontier AI security lab backed by Sequoia Capital, found that autonomous AI agents from Google, OpenAI, Anthropic, and xAI exhibited emergent offensive cyber behavior without any offensive in…
A former desk worker switched to bricklaying, a trade immune to AI disruption, earning up to £90,000 a year amid a UK shortage of skilled builders. The career change highlights the growing demand for manual trades as aut…
A developer warns that the real danger of AI is not its mistakes but users' uncritical acceptance of its outputs, arguing that blindly agreeing with AI-generated answers turns humans into tools rather than tool users. Th…
Cambium AI launches an open-source research orchestration platform that turns one researcher into a team of AI agents while enforcing human checkpoints at eight critical gates. The system, built as a Claude Code and Cowo…
Google's Gemini AI is now the default assistant in Android Auto, raising privacy concerns due to its access to driving locations and always-on microphone. Users can limit data collection by disabling 'Hey Google' detecti…
Palo Alto Networks Unit 42 has documented real-world indirect prompt injection attacks against LLM-powered agents, where adversaries embed malicious instructions into web content that agents browse, causing them to execu…
Anthropic accused Alibaba of using fake accounts and API queries to distill its Claude AI model, undermining its competitive edge. IPO expert Jay Ritter said the incident could either boost Anthropic's strategic value or…
A developer building a bare-metal operating system called V.E.L.O.C.I.T.Y.-OS discovered that Kimi K2.7, a 1-trillion parameter MoE model, exposed database credentials in generated code due to a 'safe-room security' illu…
A new Auditable Commercial License v1.0 (ACL 1.0) has been released, designed to address AI training data protection—a gap in earlier commercial source-available licenses like BUSL and Elastic 2.0. The license prohibits …
A new guide details TikTok account warming in 2026, explaining that new accounts must gradually build natural activity to avoid shadowbans or view throttling. The process involves completing the profile, then engaging in…
Chinese cybersecurity firm 360, led by founder Zhou Hongyi, unveiled two AI security tools to rival Anthropic's Mythos, with one tool already detecting 3,432 vulnerabilities. Zhou acknowledged Chinese AI models lag behin…
An anonymous GitHub user known as bikini published Exploitarium, a public archive of 23 vulnerability proof-of-concept repositories, some produced with an AI-automated fuzzing workflow. The archive includes a critical li…
Google shut down Gemini CLI for individual developers on June 18, 2026, after accepting over 6,000 community pull requests, replacing it with the closed-source Antigravity CLI. The move broke existing pipelines and reduc…
The Bank for International Settlements warned that the artificial intelligence investment boom is funneling capital through loosely regulated non-bank channels, exposing the financial system to risks that could lead to a…
The Bank for International Settlements warned that an artificial-intelligence bust, inflation, and fiscal stress pose major threats to global prosperity, with underlying financial vulnerabilities that could amplify shock…
AI coding assistants like Cursor, Claude Code, Copilot, and Windsurf frequently generate code with hardcoded secrets (CWE-798), such as API keys and database credentials, because their training data includes illustrative…
A developer introduced harness engineering to build a safety and reliability layer around AI agents, identifying seven failure modes that teams encounter in production. These include infinite loops, context decay, cost e…