EU AI Act August 2 Enforcement: GPAI Fines Go Live in 19 Days
EU AI Act enforcement goes live August 2 — GPAI providers face fines up to 3% global turnover. Here's what agentic AI teams must do in the next 19 days.
EU AI Act enforcement goes live August 2 — GPAI providers face fines up to 3% global turnover. Here's what agentic AI teams must do in the next 19 days.
The New Stack (GitLab-sponsored) examines why enterprise AI agent teams need MCP's governance and security controls to scale safely in production.
Gartner and IDC argue multi-agent systems need event-driven architecture for real-time context — Solace outlines the enterprise pub/sub playbook.
Anthropic's Project Glasswing scales to 150 orgs across 15 countries using Claude Mythos to find zero-days in critical infrastructure at scale.
Claude Sonnet 5 shipped June 30 — Anthropic's most agentic Sonnet with 1M context, near-Opus performance, and adaptive thinking enabled by default.
Two new AI agent attack patterns: Friendly Fire hijacks defensive code scanning, HalluSquatting plants malware at hallucinated package names. Both confirmed.
Wiz Research's GhostApproval exposes symlink bypass in 6 AI coding tools including Claude Code, Cursor, and Amazon Q — only three have shipped fixes.
OpenClaw 2026.7.1 stable lands with Claude Sonnet 5, Mythos 5, ClawRouter routing, native mobile apps, and conversational Crestodian onboarding.
Claude Cowork now lets agents draft emails, manage calendars, and update SharePoint files — making it a direct agentic player in M365 enterprise workflows.
Cloudflare Precursor replaces CAPTCHAs with session-long behavioral monitoring — now the primary defense as AI bots top 57% of web traffic.
Sysdig's full technical breakdown of JADEPUFFER — the first autonomous AI ransomware agent that exploited Langflow to run an end-to-end attack without human input.
JADEPUFFER's encryption key is randomly generated and immediately discarded — paying the ransom is pointless. Only offline backups can save victims.
OpenClaw beta.6 ships the Logbook plugin for AI-powered private work journals, plus Claude Sonnet 5 and Mythos 5 model support.
Claude Code July 2026 ships a sandboxed in-app browser, an interactive /doctor fixer, per-agent cost budgets, and +50% rate limits through July 19.
GPT-5.6 Sol went public July 9—OpenAI's agentic flagship with ultra parallel sub-agents. Usage limits temporarily lifted July 12 for paid plans.
Illinois SB 315 signed July 6: first US state requiring annual third-party AI safety audits for frontier labs. Applies from 2028, audit summaries public.
OpenClaw 2026.7.1-beta.6 ships iOS spoken replies with TTS, Claude Sonnet 5 & Mythos 5 models, Logbook journal plugin, and crash-loop recovery.
Two CVSS 9.8 zero-click RCE flaws in Cursor AI IDE (now patched in v3.0) show how prompt injection via MCP or web results can escape sandboxes entirely.
Gemini 3.5 Pro targets July 17 GA with a 2M token context and Deep Think mode — Google's flagship model for agentic workflows may arrive this week.
OpenAI's GPT-5.6 Sol Ultra solved a 50-year-old math conjecture in under an hour using 64 parallel sub-agents — a landmark moment for agentic AI research.
Claude Fable 5 free promo access ends tonight July 12 at 11:59 PM PT—after that, it's $10/M input, $50/M output. Pipeline operators: act now.
OpenAI's GPT-Live-1 brings full-duplex voice that listens while speaking and silently delegates complex agentic tasks to GPT-5.5 in the background—a new voice-meets-agent paradigm.
Anthropic ships native sandboxed browser in Claude Code desktop — Claude can now browse, click, and verify live sites directly in-app without switching tabs.
GitHub Copilot in VS Code ships agentic browser tools GA, parallel agent sessions, per-session cost visibility, and GPT-5.6 model support.
OpenClaw 2026.7.1-beta.5 lands with Crestodian AI onboarding, ClawRouter routing, openclaw attach, offline mobile chat, and Apple Watch voice turns.