Breaking news, AGI updates, and analysis of the intelligence revolution.
📰 News AMI Labs CEO Alexandre LeBrun calls 'superintelligence' a word with no definition and warns that today's robots have no brain. The refusal to chase the AGI label is the sharpest critique of the frontier-lab race yet.
📰 News Only 5% of enterprises fully trust their automated agent evaluations. Half have shipped agents that passed evals then failed customers. Two-thirds are removing the human check anyway.
📰 News Kimi K3 is the first open 3T-class model, with native vision, 1M-token context, and benchmark scores within striking distance of Claude Fable 5 and GPT 5.6 Sol.
📰 News Kioxia's market cap halved from its June 2026 peak as investors question whether AI chip demand is sustainable. A 660% YTD surge, 74% operating margins, and a 12% single-day drop — the AI chip boom's first stress test.
📰 News Z.AI is approaching $1B in ARR — the first Chinese AI firm to get there. Revenue grew 132% to RMB 724M in 2025, the API business grew 60x, and the models are free to download. But the company is still lossmaking and state-dependent.
📰 News Albanese's AI speech was the strongest copyright protection pledge from any world leader — but the enforcement mechanism is legislation that won't reach parliament until early 2027, against companies that have already set their own terms for every government that tried to regulate them.
📰 News Grok Build is 1.3M lines of Rust on GitHub. The trust question isn't about xAI alone — it's about whether any AI agent with filesystem access can be trusted.
📰 News Hyundai's buyout of SoftBank's Boston Dynamics stake for ~$335M gives it 100% ownership of the Atlas humanoid robot maker, accelerating its physical AI strategy and potential US IPO.
📰 News Japan's 27,500-chip deal with Nvidia for a sovereign robotics AI model is the latest move in a global sovereign AI arms race — and a bet that factory automation is the answer to a shrinking population.
📰 News The T3000 delivers near-T5000 inference at half the size and power, targeting humanoid robot deployments from Boston Dynamics to 1X. Cosmos 3 Edge brings 4B-parameter world models on-device.
📰 News OpenAI trained an AI super-hacker called GPT-Red to attack its own models. It succeeds where human red-teamers fail — and GPT-5.6 is the safer for it. But the asymmetry between attacker and defender is now an arms race inside one company.
📰 News €9.3 billion in a single quarter, a raised full-year forecast, and 30% more EUV machines planned for next year. ASML says AI demand is 'extremely strong' — and they're the only company that can supply it.