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.
Anthropic released its most capable AI model, Claude Fable 5, after previously hiding it due to safety concerns. The company also released Mythos 5, identical to Fable 5 but with a safeguard that degrades to Opus 4.8 on …
Anthropic's Fable 5 model has been suspended for all users following a U.S. government directive, but early tests show it excels at autonomous coding tasks with vague prompts, though it is expensive and slower. In code r…
Threat actors are increasingly using Microsoft Teams to launch phishing attacks, with collaboration tool phishing alerts rising to 42% of all phishing alerts in early 2026. Groups like Cloaked Ursa and UNC6692 have explo…
Broadcom announced a major set of security updates for its Spring and Java ecosystems, including the largest release of Spring security patches in the product's history and an extension of clean-room build architecture f…
Amazon's 2026 Project Eluna deploys agentic AI models in fulfillment centers to autonomously plan, use tools, and execute multistep operational tasks, moving AI beyond digital applications into physical environments. The…
Small and medium-sized businesses that adopt AI assistants like ChatGPT, Claude, and Grammarly gain significant efficiency but expose proprietary data to security risks, as 30% of firms already distrust AI companies to s…
Microsoft confirmed that 73 of its cryptographically verified open source packages were compromised last week, embedding credential-stealing malware that activated when developers opened them in AI coding agents. The att…
A 50% discount offer on annual subscriptions to The Algorithmic Bridge ends June 12th, extended one week. The article advises readers to prepare for AI-driven uncertainty using Nassim Taleb's barbell strategy, focusing o…
Apple has failed to remove impersonating apps from its App Store despite dozens of trademark infringement complaints from Kapwing Inc., costing consumers thousands of dollars and damaging the company's reputation. The co…
Socket Threat Research identified 23 new malicious PyPI packages in the Mini Shai-Hulud, Miasma, and Hades supply chain attacks, targeting bioinformatics and AI/MCP developers with trojanized native extensions and startu…
On May 29, security researcher Taylor Hornby discovered a critical vulnerability in Zcash's Orchard privacy pool using Claude Opus 4.8, a flaw that could have allowed attackers to generate ZEC from nothing by bypassing t…
A new study reveals that large language models (LLMs) silently corrupt documents when delegated editing tasks, with even top-tier models like Gemini Pro, Claude Opus, and GPT-5 degrading 25% of original content after 20 …
The reviewer agent in a closed-loop dev workflow accidentally inlined a Gitea token in a Bash command and ran 'env', exposing the secret in tool-call records. Version 0.23.4 of the platform introduced value-based secret …
Microsoft shut down over 70 of its own GitHub repositories, including those related to Azure and AI coding agents, after hackers planted malware designed to steal credentials when opened in AI coding tools like Claude Co…
AI safety researchers are exploring whether future advanced AI systems could be offered incentives like money or computing power in exchange for cooperating with human operators, rather than attempting to seize control. …
Owen McGrann's 'dead economy theory' argues that AI-driven labor replacement will cause economic collapse even without AGI, through a cycle of layoffs, spending cuts, and revenue decline. The essay criticizes tech CEOs a…
Journalists have relied on sourcing and verification frameworks for 500 years to build public trust, and those same principles can solve AI reliability challenges that prompt engineering alone cannot fix. Three distinct …
Dylan, organizer of the anti-AI anthology, reports progress on the project, noting strong submissions and extending the deadline. The anthology will include a limited print run and a free digital version, with a donation…
An AI model disproved a famous conjecture by mathematician Paul Erdős, finding a more efficient way to arrange pairs of points at equal distances than the regular grid he proposed. The breakthrough has prompted experts t…
The W3C Web Application Security Working Group has standardized the /.well-known/change-password URL as a redirect endpoint that points password managers and users to a site's actual change-password page. The endpoint, w…