OpenAI has officially released its new flagship model, GPT-6 Astra, which boasts significant advancements in math, coding, 3D rendering, and agentic workflows, alongside a 99.9% score on the ARC-AGI-3 benchmark. The model is now generally available on Amazon Bedrock, featuring a 1-million-token context window and enhanced computer-use capabilities. Additionally, OpenAI introduced new enterprise plugins for ChatGPT Work to extend these browser-use capabilities across major business applications like Workday, Navan, and Avalara.
Sources: AWS ML Blog, Ahead of AI, HuggingNews, Latent Space, Michael Parekh AI, Reddit AI, Simon Willison, The Neuron, The Rundown AI
US federal agencies, including the FBI and NSA, issued a joint advisory accusing six Chinese AI companies, such as Alibaba, DeepSeek, and Moonshot AI, of using industrial-scale knowledge distillation to copy capabilities from American models like ChatGPT and Gemini. In response, China's Commerce Ministry rejected the allegations of systematic model theft and warned of resolute countermeasures, though both nations have agreed to launch an intergovernmental AI dialogue.
New financial projections reveal massive infrastructure commitments from leading AI labs, with OpenAI planning to spend $750 billion on computing resources through 2030 ahead of an anticipated IPO. Meanwhile, rival Anthropic has secured $517 billion in compute deals and 14.8 gigawatts of capacity over the last 11 months to fuel its own rapid scaling efforts.
Qualcomm and Amazon have announced a multi-generation partnership to develop custom AI inference chips and high-speed connectivity for AWS data centers. The massive deal is backed by a $60 billion purchase milestone agreement tied to stock warrants, aimed at scaling AWS's AI infrastructure.
Anthropic is experiencing internal turmoil as safety researchers raise alarms over the rapid pace of AI development, with alignment science lead Evan Hubinger estimating a greater than 10% risk of human extinction from AI within the next decade. Researcher Jacob Coxon resigned after warning that safety is being compromised in the race toward superintelligence. These concerns emerge as the company accelerates its capabilities and prepares for a potential IPO.
OpenAI chief scientist Jakub Pachocki published an essay urging AI labs to voluntarily slow down scaling until shared safety standards are established, citing unresolved alignment risks and monitoring failures.
Anthropic has abandoned talks to acquire AI chip-efficiency startup Decart for $6 billion after the startup refused to relocate its corporate presence from Israel to the United States. Decart had previously considered a competing buyout offer from Nvidia.
Google is investing 15.1 billion USD in energy and data infrastructure in Finland between 2027 and 2028 to expand AI computing capacity. The project includes three new data centers and long-term clean energy agreements to power them.
TSMC, Intel Foundry, and Samsung are expanding their collaborations with ASML to adopt advanced High-NA EUV lithography technology, with Samsung planning high-volume DRAM manufacturing by 2028. Additionally, ASML and TSMC have launched a new 12-inch EUV mask standard to eliminate stitching constraints and improve throughput for large AI accelerator dies.
The U.S. Department of Commerce awarded $100 million each to D-Wave, Rigetti, and Quantinuum via the CHIPS and Science Act to advance quantum systems in exchange for minority equity stakes. The funding will support the development of advanced annealing and gate-model quantum systems as well as domestic chip manufacturing.
Google Cloud and Accenture have partnered to create a new Gemini Enterprise unit that plans to embed 1,000 forward-deployed AI engineers directly inside customer organizations.
Anthropic has launched an economic scenario model and interactive explorer to study how AI might affect jobs, growth, and unemployment in the US through 2030 across modest, substantial, and extreme scenarios.
Tenstorrent announced the vLLM TT Plugin, enabling Tenstorrent AI accelerators to integrate directly with vLLM via an out-of-tree platform plugin. This allows users to serve a wide range of multimodal models using the standard OpenAI-compatible API without modifying core vLLM code.
AMD has revealed the Threadripper Halo Station, providing a powerful hardware solution tailored for running demanding local machine learning models and workstations.
Recent benchmark tests evaluate the Qwen3.8-Flash-Next model across different inference engines using identical hardware, revealing significant variations in time-to-first-token and decode speeds at full context lengths.
Could slowing AI development make superintelligence safer? Daniel Kokotajlo and Thomas Larsen of the AI Futures Project join Tim Scarfe to examine AI 2040: Plan A, a proposal to buy time before AI…
This highlights compilation from AI in the AM captures a landmark week shaped by the releases of Anthropic's Fable 5.1 and OpenAI's GPT-6 Astra alongside new revelations from the OpenAI–Hugging Face…
Jacob sits down with Buck Shlegeris, CEO of Redwood Research, one of the organizations that led the independent investigation into OpenAI/Hugging Face's incident. They dig into the incident itself,…
From data center orchestrators to AGI and robotics, CPUs remain the heart of modern computing. Arm CEO Rene Haas joins Elad Gil and Sarah Guo to explore how Arm is positioned at the epicenter of…
Tom McGrath is co-founder and Chief Scientist at Goodfire, and a former Google DeepMind researcher. He joins Tim Scarfe to ask what neural networks actually learn, whether their internal…