Breaking news, AGI updates, and analysis of the intelligence revolution.
📰 News Huawei Ascend 910C chips completed full-parameter post-training of DeepSeek-V4-Pro's 1.6 trillion parameters across 1,000+ chips, 1,500 iterations, zero errors — China's first domestic-training milestone.
📰 News Anthropic paused Claude Agent SDK billing changes hours before they were set to take effect. Power users who would have faced major cost increases get a reprieve — but for how long?
📰 News A federal judge forced OpenAI to hand over 20 million conversations users thought they deleted. There is no legal privilege for what you tell ChatGPT.
📰 News Amazon MGM shelved Luca Guadagnino's nearly-finished Altman biopic months after a $50B OpenAI deal. The timing is not subtle.
📰 News The UK's AI and data regulator has lost its leader — the first ICO commissioner to ever resign mid-term, amid a workplace investigation and pending government reforms.
📰 News Barret Zoph left OpenAI in 2024, co-founded Mira Murati's rival lab, came back in January 2026, and is now gone again. OpenAI's enterprise push just lost its leader.
📰 News Weibo's VibeThinker-3B matches models 224x its size on math benchmarks. Real-world testing reveals gaps. The benchmark vs reality tension is the real story.
📰 News Waymo's sixth recall pulls 4,000 robotaxis from highways after 13 construction zone incidents. The pattern is clear: aggressive expansion is outpacing edge-case competence.
📰 News Kevin Hassett compared future AI safety rules to FDA drug approvals. CAISI has 40 evaluations done and 30 staff. The hands-off era is ending — and NZ has no seat at the table.
📰 News The EFF filed an amicus brief on 18 June 2026 arguing the Trump administration's export controls on Anthropic's Mythos and Fable models are unconstitutional retaliation. The brief claims Anthropic was singled out after refusing to let the Pentagon use its models for autonomous weapons and surveillance, that other AI models with similar capabilities face no equivalent restrictions, and that the First Amendment precedent from the 1990s encryption wars — 'code is speech' — applies directly to AI model exports.
📰 News DeepMind's AI Control Roadmap treats its own agents as potential insider threats, using MITRE ATT&CK-based threat modelling and trusted AI supervisors to monitor agent behaviour in real time.
📰 News The man whose name is in 'Generative Pre-trained Transformer' is moving across the street to the company that put the GPT in those letters.