New research tested GPT-4.5, Claude Opus 4, Gemini 2.5, Llama 4, and Mistral Large across 1,000 real-world fact-check claims. They only agreed on 33% of verdicts - raising serious questions about using LLMs as arbiters of truth.
Simon Willison argues that both Anthropic (Claude Code) and OpenAI (Codex, ChatGPT) have finally nailed product-market fit with AI coding tools and agent workflows, marking a turning point for the industry. 300+ points on HN.
Weekly roundup covers a Claude-powered security plugin, Azure privilege escalation, Kali365 MFA bypass technique, FIFA scam campaigns, and 15+ other active threats. Exposed infrastructure and recycled social engineering remain the theme.
The world's largest cruise operator confirmed a breach claimed by ShinyHunters in April 2026. Nearly 6 million customer records exposed. Another reminder that hospitality remains a prime target.
Meta is rolling out paid subscription plans for WhatsApp, Instagram, and Facebook worldwide under its 'Meta One' brand. AI-powered features, creator tools, and business tiers are part of the broader subscription push.
Cognition (makers of Devin) hit $492M annualized revenue and more than doubled its valuation in eight months. The $1B raise signals continued massive appetite for AI coding agent companies.