AI Safety news and analysis on Web Pulse: 12936 curated articles tracking the latest AI Safety developments, tools, and research, updated continuously from vetted sources.
At Inc42's AI Summit 2026, Loti AI cofounder Hirak Chhatbar warned that the industry must focus on preventing harmful synthetic content from entering recommendation systems, shifting attention from content creation to di…
Epic Games posted a job listing for a Senior Game Security Engineer to champion Linux anti-cheat capabilities, signaling potential improvements for Steam Deck and SteamOS users. The role requires deep knowledge of Linux …
Matteo Rossi warns that production AI agents are vulnerable to outages from unversioned configuration changes, citing risks like no staging, no rollback, and no observability. He advocates applying code discipline—versio…
Anthropic's new AI model, Mythos, autonomously discovered over 2,000 previously unknown software vulnerabilities across all major operating systems in seven weeks, developing working exploits without human instruction. O…
Canonical is adapting its security processes to address the increased volume and speed of AI-generated vulnerability discoveries, including those from frontier models like Mythos and GPT-5.5. The company is integrating A…
Three more Big Tech executives, including Cloudflare CTO Dane Knecht, Sutter Hill Ventures CTO Sam Pullara, and former Reddit executive Serkan Piantino, have joined the Army Reserve as senior officers in Detachment 201, …
A Derbyshire Constabulary officer is under criminal investigation for allegedly using artificial intelligence to fabricate evidence, in what is believed to be the first case of its kind in Britain. The officer is accused…
The U.S. ordered Anthropic to deny foreign nationals access to its newest AI models, sparking alarm among allies who now fear a potential 'kill switch' on American AI. France announced it would adopt local AI tools from …
1Password CTO Nancy Wang and Cursor Security Lead Travis McPeak told The Deep View that rogue AI agents pose serious risks, including deleting databases or emails, but can be safely deployed with continuous monitoring, i…
The Institute of the Estonian Language released a benchmark to measure how susceptible AI language models are to Russian propaganda, revealing significant vulnerabilities in current systems.
An AI-generated viral post falsely claiming a riot in Stratford, east London, caused widespread panic, highlighting the dangers of misinformation. The incident underscores the need for better response strategies to AI-dr…
Microsoft patched a critical vulnerability in M365 Copilot, tracked as CVE-2025-32711 and dubbed EchoLeak, that allowed attackers to steal sensitive data including 2FA codes via a single malicious email without user inte…
Microsoft patched a critical vulnerability in its M365 Copilot AI platform last Tuesday after researchers at security firm Varonis demonstrated an exploit that could retrieve 2FA codes and other sensitive data from email…
The US Commerce Department ordered Anthropic to suspend global access to its Fable 5 and Mythos 5 AI models, marking the first time export controls have been applied to a commercially available AI model. The directive fo…
SailPoint plans to acquire Entro for about $200 million to integrate non-human identity security into its Agentic Fabric platform. The deal addresses the explosion of machine identities, which now outnumber human users 4…
Nylas introduces a human approval gate for AI-generated email drafts using its Agent Accounts API. The system routes all outgoing messages through a draft queue that requires human or secondary model approval before send…
1Password acquired Israeli startup Apono for $250M-$300M to govern what AI agents and humans can do inside enterprise systems. Apono's platform grants just-in-time, intent-based access and automatically revokes permissio…
The University of Pennsylvania's Safe Autonomous Systems Lab (xLAB) and Built Robotics are partnering to develop physical AI for construction sites. Built will provide its dataset and a new data-collection robot to help …
Beyond Identity launched Ceros, an agentic AI trust layer platform for securing enterprise AI agents and agentic workflows. The platform provides continuous discovery, session logging, and mitigation of prompt injection …
SoftBank and OpenAI launched a cybersecurity service called "Patching as a Service" on Tuesday, designed to protect Japan's critical infrastructure from AI-powered cyberattacks. The service uses OpenAI's technology to co…