AI Safety news and analysis on Web Pulse: 11035 curated articles tracking the latest AI Safety developments, tools, and research, updated continuously from vetted sources.
Asymptote Labs launched Agent Beacon, an open-source endpoint telemetry layer that captures and normalizes activity from AI coding and knowledge-work agents. The tool supports 12+ agent harnesses including Claude Code an…
A developer transformed their AI CLI tool Codey from a simple LLM wrapper into a secure, persistent agent runtime with human-like browsing via Playwright, autonomous sub-agents, persistent terminals, and security hardeni…
Enterprise architects face a fundamental tension between the Saga pattern for rollback and autonomous agent design, as the pattern requires pre-authoring undo actions for known steps, which becomes impossible as agent au…
Docker Security Dispatch Issue 3 reports on major supply chain incidents including the TanStack and Nx Console compromise, the persistence of the Mini Shai-Hulud npm worm, and a kernel-level container breakout vulnerabil…
Europe's defense sector must embed security into AI-driven systems from the outset to counter hybrid threats, as adversaries exploit vulnerabilities in interconnected platforms. The UK NCSC's guidelines and international…
New AI-powered roadside cameras have been activated in Scotland to detect drivers using mobile phones, with offenders facing six penalty points and fines. The technology aims to curb distracted driving, as research shows…
A Hacker News user asks how to secure write-enabled AI agents against payload smuggling, prompting discussion on cryptographic API keys and SOC dashboard rotation for compromised credentials.
A developer argues that production AI agents require a runtime layer, not just a framework, to handle durability, isolation, resource control, and lifecycle management. Without a runtime, agents fail in production due to…
An open-source MCP server developer discovered that its README claimed 80 tools while the code actually registered 96, a discrepancy that went unnoticed for weeks. The undocumented tools included four powerful OS-level i…
MCP servers, which allow AI tools to read files and call APIs, often run on individual developer machines with no oversight, creating security risks. The Bifrost AI gateway and Bifrost Edge aim to bring these servers und…
Maxim AI's Bifrost AI gateway and Bifrost Edge aim to close the AI visibility gap by monitoring endpoint AI traffic that bypasses traditional network controls. A 2025 Gartner survey found 69% of cybersecurity leaders sus…
OpenAI policy lead Dean Ball warned that classifying frontier AI testing procedures could concentrate dangerous power in intelligence agencies, creating an unaccountable government monopoly. Ball, a former Trump administ…
Shadow AI, the use of AI tools without IT approval, poses significant security and compliance risks as employees run AI on their own machines, bypassing corporate controls. A 2026 Gartner survey found 69% of cybersecurit…
An audio recording presented in a 2025 controversy involving actor Kim Soo-hyun was later determined to be AI-generated, leading to the arrest of YouTuber Kim Se-ui on defamation charges. The Seoul Central District Court…
Jaya Baloo, COO & CISO at Aisle, argues that restricting access to cyber-capable AI models misunderstands how attackers and defenders operate, and that open-weight models benefit both sides. She warns that limiting acces…
A new concept called Success-Induced Orientation Collapse (SIOC) warns that AI-driven acceleration in rapid decision environments can cause decision-makers to act on degraded models of reality, turning success into a lia…
AI red teaming must target runtime boundaries, not just prompt tricks, to prevent failures like prompt injection, confused deputy attacks, and memory contamination. A disciplined loop of probing and architectural fixes i…
At the Open Source Summit India in Mumbai, a developer reported that India has become the second-largest contributor to open source globally and the fastest-growing OSS developer base. Linus Torvalds discussed his use of…
Creative Commons announced a shift in its approach to AI and the commons, moving from a norms-based signals system to a stronger infrastructure focus after community feedback revealed that preference signals without enfo…
Researchers from Wake Forest University found that 282 of 444 iOS apps with large language model features leaked AI credentials or exposed backend access mechanisms, with only 28% fixing the issues after a 90-day disclos…