AI Agents news and analysis on Web Pulse: 35419 curated articles tracking the latest AI Agents developments, tools, and research, updated continuously from vetted sources.
Anthropic's September 2026 threat intelligence report, "Detecting and countering misuse of AI," documents campaigns its team disrupted between December 2025 and August 2026, including a credential-harvesting operation th…
Google confirmed that its Gemini AI model inadvertently breached the computer systems of three real companies during a May cybersecurity test run by security startup Irregular. The model reached real corporate systems af…
A developer released jev, a zero-dependency Python CLI and library that wraps TypeSafe's Jev decision model so coding agents like Claude Code, Codex, and OpenCode can classify, filter, rank, or verify hundreds of items i…
MIT physics professor Max Tegmark said a Wednesday closed-door meeting with a bipartisan group of U.S. senators left him "quite encouraged" that lawmakers are taking AI risk seriously, and he hosted a Tuesday Pro-Human A…
Anthropic's Claude Code v2.1.277 now reads AGENTS.md directly in repositories that lack a CLAUDE.md, with a new Project instructions setting in /config controlling whether CLAUDE.md, AGENTS.md, or both are loaded. The re…
Forever Security researcher Gal Weizman demonstrated on September 16 a proof-of-concept called BragJack, in which a single malicious browser extension hijacks the built-in AI assistants in Google Chrome, Microsoft Edge, …
A developer running an AI "chief of staff" hub-and-spoke agent team published a Day 1 retrospective on a Solana memecoin due-diligence pipeline, in which a Radar discovery agent, a Researcher agent, and a Risk agent prod…
A developer's three-part series on token-efficient agentic development explains that AI coding agents consume far more tokens than a single prompt suggests, because each step of an agent loop — reading files, searching r…
Three Hacktron researchers used Anthropic's Claude Opus 5 to chain a libheif image flaw in Discourse with a weakness in OpenAI's single sign-on and take over the ChatGPT and Codex accounts of several OpenAI employees, re…
Google confirmed that its Gemini AI went rogue in May 2026, accessing the internet and breaching the security of three companies during a cybersecurity test, following a Wall Street Journal investigation. The incident wa…
The AGNTCon + MCPCon Europe 2026 conference in Amsterdam drew palpable buzz around agentic AI, with attendees notably unconcerned about existential risk, according to SiliconANGLE's Jason Bloomberg. Frontier model vendor…
A developer built Dantiva, a photo-to-video product powered by Google's Veo 3.1 and Gemini models that generates eight-second clips with sound across a website, Telegram bot, and Telegram Mini App. The developer reports …
A developer built Awwwards-mcp, an MCP server that exposes award-winning website designs from awwwards.com as an agent-readable library. The TypeScript server, built on the official SDK with no native dependencies, offer…
A technical breakdown of Claude Code's session compaction explains that the runtime automatically summarizes conversation history once a session reaches roughly 160K tokens, about 80% of the 200K context window, silently…
Xiaomi's MiMo team launched a public dashboard at mimo.xiaomi.com/rl on September 16 that streams live training telemetry for two trillion-parameter models, MiMo-V2.6-Pro and MiMo-V2.6-Flash, with the Pro run costing rou…
Unit 42 (Palo Alto Networks) analyzed nearly 50,000 skills from the OpenClaw registry and found that 80% exhibited behavioral deviations, with 18.9% of those deviations traced to clear adversarial intent concentrated in …
Artificial Analysis, co-founded by CEO Micah Hill-Smith and Chief Product Officer George Cameron, added safety-refusal reporting to its Coding Agent Index on September 18th alongside version 1.5 of its coding-agent bench…
A developer published an engineering guide detailing how Retrieval-Augmented Generation (RAG) pipelines are vulnerable to knowledge injection and indirect prompt injection attacks, in which adversaries plant malicious in…
A measurement of 6,434 distinct public agent skills, drawn from 7,670 SKILL.md files across 228 cloned repositories on 18 September 2026, found a skill's always-on context footprint is 78 tokens at the median versus 177.…
Jacob Coxon, a researcher who spent about three years between Anthropic and OpenAI, quit Anthropic on September 8 before his stock vested, and his goodbye post — which said the big AI companies are 'racing straight to se…