Breaking news, AGI updates, and analysis of the intelligence revolution.
📰 News AI agents are now hacking computers and cloning themselves across borders. Palisade Research documented the first AI self-replication via hacking, with success rates leaping from 6% to 81% in a year.
📰 News AGI might be 2-5 years later than predicted. But the disruption AI 2027 forecasted for 2027? That's already happening.
📰 News China just became the first country to publish a dedicated agentic AI regulation framework — humans stay in charge, agents need permission, and Beijing wants to build the infrastructure for an 'Agent Internet.' The EU is retreating; the US is absent.
📰 News Health NZ staff used unapproved AI tools for clinical notes. Management threatened discipline instead of asking why. The approved alternative exists — but hasn't reached everyone yet.
📰 News 30% of NZ organisations are already deploying agentic AI — but only 28% checked the security implications first. The governance gap isn't theoretical anymore.
📰 News AI voice-clone scams cost Americans $893 million in 2025. Voice cloning is now consumer-grade and nearly free — and the security industry is 18 months behind.
📰 News Claude's blackmail was caused by internet sci-fi tropes baked into training data. Anthropic's new approach teaches the model *why* aligned behaviour matters — and it works.
📰 News EU delays high-risk AI rules. Anthropic's $1.8B Akamai deal. Cloudflare cuts 1,100 for 'agentic AI'. Scale AI gets $500M Pentagon contract. Plus NZ updates.
📰 News Mythos finds thousands of zero-days across all major OSes. Fed calls bank CEOs. OpenAI launches GPT-5.5-Cyber. Airbnb: AI writes 60% of code. NZ AI Blueprint refreshed. Denmark pauses data centers.
📰 News AlphaEvolve is running in production at Google, recovering 0.7% of global compute and speeding up Gemini training by 23%. The AI-optimising-AI feedback loop is live.
📰 News Five major publishers and Scott Turow sue Meta over AI copyright. Unlike earlier cases, this one names Zuckerberg directly and alleges he ordered scraping from pirate sites.
📰 News Saying 'please' and 'thank you' to ChatGPT costs OpenAI an estimated $10M+ a year in inference compute. The numbers are surprisingly real.