MIT's EnergAIzer Estimates AI Power Use in Seconds
MIT and MIT-IBM Watson AI Lab researchers have developed EnergAIzer, a method for estimating the power draw of AI workloads in seconds. The tool could help data center operators, model developers, and chip designers reduce wasted energy as AI demand grows.
Microsoft and OpenAI Redraw the Rules of Their AI Partnership
Microsoft and OpenAI have amended their partnership, giving both companies more room to maneuver while preserving Azure's central role. The deal clarifies cloud access, IP licensing, revenue sharing, and the companies' long-term AI infrastructure plans.
China Orders Meta to Unwind Its Manus AI Acquisition
China has ordered Meta to reverse its acquisition of Manus, a Singapore-based AI startup founded by Chinese engineers. The decision could unsettle Chinese AI founders seeking foreign capital and deepen the political risk around cross-border AI deals.
Google Lines Up $40B Anthropic Bet as AI Compute Race Intensifies
Google is reportedly preparing up to $40 billion in cash and compute support for Anthropic, including $10 billion now and another $30 billion tied to performance targets. The deal deepens Anthropic's dependence on hyperscale AI infrastructure.
OpenAI Opens GPT-5.5 Bio Bug Bounty to Test Frontier AI Safety Controls
OpenAI is inviting vetted red-teamers, security researchers, and biosecurity specialists to test whether GPT-5.5 can be universally jailbroken on five biology safety questions. The private bounty offers $25,000 for the first true universal jailbreak.
Anthropic Brings Claude Connectors Into Travel, Shopping, Food, and Daily Tasks
Anthropic is expanding Claude connectors beyond workplace apps into consumer services including AllTrails, Instacart, Audible, TripAdvisor, Uber, Spotify, Resy, and TurboTax. The update turns Claude into a more practical everyday assistant, while raising the stakes for consent, ranking, and data access.
AWS Makes Frontier Agents Its Next Big Enterprise AI Bet
AWS is pitching frontier agents as the next phase of enterprise AI: systems that can work autonomously, at scale, and across long-running goals. The company is using Kiro, DevOps agents, security agents, and sustainability data projects to show where the model could land first.
Microsoft Commits $18 Billion to Australia's AI Infrastructure Push
Microsoft plans to spend $18 billion in Australia through 2029, expanding Azure capacity, cybersecurity partnerships, AI safety work, and workforce training. The deal is part of a broader global race to secure compute, policy alignment, and national AI capability.
Anthropic Gives Claude Managed Agents Built-In Memory for Long-Running Enterprise Work
Anthropic has launched built-in memory for Claude Managed Agents in public beta, giving agents a filesystem-based way to learn across sessions. The update targets enterprise teams that want agentic systems to retain lessons, reduce repeated mistakes, and stay auditable in production.
OpenAI Launches GPT-5.5 for Real Work Across ChatGPT, Codex, and Enterprise Teams
OpenAI has launched GPT-5.5 as a new flagship model aimed less at chatbot novelty and more at sustained work across coding, research, data analysis, and agentic tool use. The release also signals a sharper push into enterprise and workflow-heavy deployments.
Google Upgrades Gemini Deep Research With a New Max Tier for Autonomous Agents
Google is expanding its Gemini Deep Research offering with a faster standard agent and a new Deep Research Max mode built for higher-quality autonomous research. The launch adds MCP support, native charts, multimodal grounding, and deeper enterprise data access to long-horizon research workflows.
OpenAI Launches ChatGPT Images 2.0 as a More Precise, Multilingual Visual Engine
OpenAI has launched ChatGPT Images 2.0 across ChatGPT, Codex, and the API, positioning image generation less as a novelty feature and more as a serious visual production system. The new model adds stronger instruction following, better text rendering, broader language support, flexible aspect ratios, and tighter links to reasoning workflows.