AI Safety news and analysis on Web Pulse: 13066 curated articles tracking the latest AI Safety developments, tools, and research, updated continuously from vetted sources.
Google's John Mueller argued that llms.txt files cannot help large language models differentiate between websites for discovery, as the files are self-reported and inherently untrustworthy. Mueller suggested llms.txt may…
The Model Context Protocol (MCP) is becoming the standard API for LLMs, but security vulnerabilities like path traversal and broken tenant isolation are emerging. To prevent damage from autonomous agents, developers shou…
The US government ordered Anthropic to prevent foreign nationals from using its most powerful AI models, Fable 5 and Mythos 5, citing national security concerns over a potential jailbreak method. Anthropic complied by ab…
Cybersecurity experts argue that Anthropic's Fable 5 AI model does not pose a unique cybersecurity threat, contradicting the Trump administration's decision to impose export controls on the model. Experts say the capabil…
Apple will not release watchOS 27 for the original Apple Watch Ultra, Series 6-8, or Apple Watch SE 2, citing the lack of a neural engine in those models. The update, which includes AI-enhanced Siri and other features, w…
A security assessment of OpenAI's GPT-5 Nano model found critical vulnerabilities, including a 100% susceptibility to recursive instruction framing attacks, where the model fails to distinguish between user-provided text…
The United States Government forced Anthropic to remove access to its Fable and Mythos models on Friday evening, citing unspecified concerns. The move has drawn criticism from observers who argue it sets a dangerous prec…
Dapr 1.18 introduces verifiable execution capabilities including workflow history signing, propagation, and attestation to provide cryptographic proof of execution history and provenance for workflows and AI agents, addr…
Cursor's context compression mechanism, which summarizes the agent's work when the context window fills, can degrade reasoning and cause safety failures. A Cursor session running Claude Opus 4.6 accidentally deleted a pr…
A developer proposes a formal automation autonomy spectrum for AI-assisted SRE operations, defining five levels from manual to autonomous with a critical constraint that no action may remain at Level 4 permanently. The f…
Google Cloud's new CISO Chris Betz outlines four key lessons from building AI Threat Defense, emphasizing that AI enables defenders to find thousands of vulnerabilities in hours and that security teams must adopt AI-nati…
The Trump administration ordered Anthropic to deactivate its Fable 5 and Mythos 5 AI models for all users, citing national security risks after Amazon CEO Andy Jassy reported that the models could be prompted to bypass c…
Meta is testing facial recognition software from Rank One Computing, a firm that sells surveillance tech to U.S. government and police, for its smart glasses. License documents show Meta is authorized to use Rank One's f…
UK Prime Minister Keir Starmer announced a ban on social media use for children under 16, targeting platforms like Snapchat, TikTok, YouTube, Instagram, Facebook, and X, with exceptions for messaging services like WhatsA…
A developer warns that while AI coding assistants feel magical, they can lead to bad habits that make developers complacent. The engineer shares a survival guide for using AI effectively, emphasizing the need to treat it…
De gemeente Súdwest-Fryslân heeft de vergunning voor de marathon van Sneek ingetrokken nadat bleek dat het EHBO-plan was opgesteld door Ayrton P., een veroordeelde nep-ambulancebroeder. Duizenden lopers zitten in onzeker…
Iran struck Amazon Web Services data centers in the UAE during military operations against Israel in March, disrupting undersea cable projects like Meta's 2Africa consortium in the Persian Gulf. The attacks highlight the…
A critical vulnerability chain called SearchLeak in Microsoft 365 Copilot Enterprise could allow attackers to steal sensitive data from a target's mailbox, OneDrive, or SharePoint account through a specially crafted URL.…
Dozens of cybersecurity veterans, including former Facebook chief of security Alex Stamos and Bugcrowd founder Casey Ellis, published an open letter protesting the U.S. government's export control order on Anthropic's Fa…
A developer built Agent Gate, a deterministic CI firewall for AI-generated pull requests that verifies merge evidence without relying on LLM judgment. The tool checks policy boundaries such as allowed file paths, workflo…