Nvidia opened an independent investigation into CEO Jensen Huang after U.S. intelligence agencies alleged that a woman he had an affair with is a Chinese spy.
Why it matters: Leadership uncertainty at Nvidia, the main supplier of AI chips, could affect supply and roadmap confidence for companies that depend on its hardware.
OpenAI describes a candidate nonperturbative lattice definition of the Standard Model and says its GPT-6 Astra model helped develop it into a detailed proof.
Why it matters: The case is a company-reported example of a frontier model contributing to original formal research. That could matter for teams assessing AI for scientific or technical analysis. The claim rests on OpenAI's own account until outside experts review the work.
Fable 5.2 completed 41.3% of real freelance projects to a professional standard in the Remote Labor Index, up from a top score of 2.5% a year earlier.
Why it matters: Clearly defined digital services that are easy to compare and swap out may face pricing pressure first, since agents can now complete a sizable share of such assignments.
Google's Gemini 4 Carbon introduces teams of agents that can work together on tasks, as major model releases keep accelerating.
Why it matters: Agent-team features built into a flagship model could change how companies split long, multi-step work between software agents and people.
Stanford HAI says global corporate AI investment has passed $1 trillion in 2026 and that job declines in AI-exposed occupations have continued into this year.
Why it matters: Companies planning AI infrastructure may face growing local resistance to data centers, and workforce effects appear even in roles where AI is used as support rather than replacement.
Google DeepMind and Google Quantum AI build a quantum program library in the Lean theorem prover, where teams of Gemini 4 agents write and prove programs.
Why it matters: Formal proof checking can replace simulation as the reward signal, letting agents search for algorithms that current computers cannot test directly.
Researchers from SII and GAIR apply recursive self-improvement to an entire frontier pre-training run from scratch and report the details.
Why it matters: Improving pre-training with agents could lift the base model that every later stage builds on, and the open release lets others test the approach.
Anthropic is spending $3.4 billion on Caterpillar natural gas turbines with a capacity of 1.7 gigawatts.
Why it matters: Anthropic taking ownership of power generation suggests that compute capacity for frontier AI is increasingly constrained by energy supply.
In a test of real work, AI coding agents needed human cleanup on nearly three out of four tasks, which shows that producing code is not the same as delivering finished work.
Why it matters: Teams measuring coding agents by code output alone may overstate productivity gains, because review and cleanup still consume human time.
Researchers from the University of Washington and the Allen Institute for AI introduce s3, a reinforcement learning method that lets agents with no preset roles learn to coordinate.
Why it matters: The result suggests coordination between agents can be learned from a shared goal rather than hand-designed in a harness, though the tests so far use very small models.
City University of Hong Kong and Tencent Jarvis Lab propose self-organized memory, in which agents decide how to summarize and link their own notes.
Why it matters: Teams building agents with memory could avoid fixed schemas, and memory written by one model was reported to work when read by another.
Microsoft Research Asia proposes LongCapHarness, which builds a harness around a frozen model rather than post-training it.
Why it matters: Before post-training a small model for long context, teams could try improving the harness around it, which the authors say is cheaper, though a strong harnessing agent is still needed.
UC Berkeley and the University of Illinois Urbana-Champaign analyze agent cost and build Agency, a layer that cuts cost without changing the model.
Why it matters: Companies running agents may find much of their bill comes from repeated context reads and extra tool calls, which can be optimized without switching models.
KAIST researchers describe memory-induced degradation and introduce WikiTrace, a memory system that links each lesson to the run it came from.
Why it matters: Agent memory can make agents worse at tasks they already solve, so keeping evidence beside each stored lesson may matter for production agents that learn from experience.
Google researchers introduce ReproBridge, which packages one successful reproduction of an ML experiment so the next agent can reuse it.
Why it matters: Reusable reproduction artifacts could make agent-run ML experiments cheaper and more consistent across models, though results come from a research benchmark.
Meta researchers present AutoRefine, a method for keeping LLM code judges calibrated as the coding agent they evaluate evolves.
Why it matters: Teams that use LLM judges to grade agent output may find them drifting as agents change, and workflow outcomes could serve as the calibration signal.
Robbyant introduces LingBot-Touch, a tactile foundation model that maps piezoresistive, magnetic and visuotactile sensors to one virtual fingertip.
Why it matters: Robotics teams could reuse one tactile model across sensor hardware instead of training an adapter for each, based on the authors' reported results.
The 0.4a0 alpha of datasette-apps lets sandboxed apps link to each other through URL fragments and reports supported URL schemes to Datasette Agent.
Why it matters: Teams building internal data tools with AI agents can have the agents generate apps that link to each other, rather than only standalone pages.
The 0.6a0 alpha release of Datasette Agent adds a prompt box to the Datasette homepage and supports hash-mentions of tables and queries.
Why it matters: Teams that use Datasette for internal data exploration get a more direct way to put an AI agent in front of their data. Because this is an alpha, its behaviour may change before a stable release.