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 generation of AI-powered toys, including the ChattyBear teddy bear, are being sold online and marketed as giving children as young as three an educational advantage. Researchers who evaluated six such toys over sev…
Only 11% of 100 commercial and publicly available AI agents assessed in a new independent report met the security bar for production use. The AI Risk Quadrant (AIRQ) report found that nearly all agents, which are deploye…
QR codes, originally designed for machine readability, are increasingly used as verification tools across the internet, requiring humans to scan them to prove they are not bots. This practice, which surged during the COV…
Anthropic analyzed 832 banned accounts that used its AI models for malicious cyber operations between March 2025 and March 2026, mapping their activities onto all 14 tactics and 482 unique techniques in the MITRE ATT&CK …
Psychologists warn that over-reliance on artificial intelligence is eroding cognitive strength by making delegation feel like self-generated thought, according to a new analysis. The first capacity lost is tolerance for …
A nurse at Erlanger Baroness in Chattanooga admitted to stealing and abusing fentanyl for months, according to a nursing-board order reported by KFF/CBS News. The hospital's AI-powered drug diversion monitoring system, S…
Anthropic has posted job listings for a new "AI & Rule of Law" team within its Anthropic Institute, offering a salary range of $295,000 to $345,000 for a San Francisco-based role. The team, led by Yale Law School Residen…
Apple will unveil iOS 27 at its WWDC keynote on June 8, featuring a major overhaul of Siri into a full conversational chatbot integrated throughout the operating system. The update places Apple at the center of a growing…
Researchers from the University of Toronto demonstrated a prototype AI-powered worm that can autonomously spread across networks, exploit vulnerabilities across multiple platforms including Linux, Windows, and IoT device…
Anthropic is hiring for a new "AI & Rule of Law" team to study how artificial intelligence impacts executive power, courts, elections, and democratic institutions. The AI company, which is moving toward a blockbuster IPO…
Anthropic cofounder Chris Olah, a self-declared atheist and billionaire AI safety researcher, sat beside Pope Leo XIV at the Vatican last week to warn that the tech industry cannot govern itself. Olah, who serves as the …
In June 2026, attackers exploited Meta's AI-powered support chatbot to take over high-profile Instagram accounts, including the Obama White House account, by convincing the AI to perform account recovery operations. The …
Anthropic expanded its Project Glasswing cybersecurity initiative to approximately 150 organizations across more than 15 countries, adding partners in healthcare, energy, communications, and technology sectors. The progr…
Netskope announced the Netskope One AI Command Center, a new addition to its Netskope One AI Security suite that combines AI discovery, risk intelligence, and autonomous response capabilities. The product aims to help se…
Attackers used prompt-injection techniques against Meta's AI-powered support assistant to hijack high-profile Instagram accounts, including the Obama White House account, by tricking the bot into changing account email a…
Pydantic, Prefect, and Hugging Face are offering a $10,000 bounty to anyone who can escape the Pydantic Monty Python sandbox and retrieve a secret stored on the hackmonty.com server. The bounty program, hosted on a honey…
Agent Threat Rules (ATR), an open detection format for AI agent security threats, has been published as YAML documents that conform to a versioned schema. The format targets attack classes including prompt injection, too…
Anthropic expanded its Project Glasswing cybersecurity program by approximately 150 new organizations, bringing the partner list to roughly 200 across more than 15 countries. Partners using the Claude Mythos Preview have…
Russian military trucks, including Ural and KAMAZ models, have been painted with black-and-white "dazzle" patterns to confuse AI-enabled machine-vision targeting used by Ukrainian drones, according to multiple reports. T…
South Korea's Science Ministry announced that the Korea Internet & Security Agency has secured access to Anthropic's cybersecurity AI model Mythos through participation in Project Glasswing, an initiative using frontier …