Clear, useful reporting on the AI models, agents, products, and research changing how we work.
Latest coverage
87 published articles
Linus Torvalds has rejected a project-wide anti-AI position for Linux, but the Sashiko dispute remains unresolved where it matters operationally—who verifies automated reviews, who receives them and whether the system reduces total maintainer work.
Sable has raised $45 million for Aidan, an agent designed to run live product demos and onboarding. The company names Notion and Decagon as production customers, but has not published the performance or cost data needed to show that the system can replace customer-facing work at scale.
Aina has raised its first $5.5 million round to launch an undisclosed interface for AI agents. Its evidence so far is Dune, a $119 Mac keypad shipped in the hundreds—not yet proof that dedicated hardware can become a broader control layer.
Raft has moved its shared workspace for people and AI agents into general availability, charging for coordination while customers keep paying model providers. Company-reported beta adoption shows interest, but close substitutes and missing productivity data narrow the claim that Raft has solved multi-agent work.
Dario Amodei and five other Anthropic employees supplied 92.5% of Public First’s itemized second-quarter contributions, giving the AI-safety super PAC campaign money while underscoring its concentrated donor base and smaller cash reserve.
OpenAI said it would restore Codex and ChatGPT Work users to 100% of their weekly usage limit after citing 9 million active users, following an Anthropic reset and a similar exchange a week earlier. The extra capacity is immediate, but neither reset changes the products’ documented standard allowances or resolves how the companies will price sustained agent workloads.
Noetra has begun a Japan-backed physical AI model program, but its 140 MW Nvidia Vera Rubin facility is not due online until June 2028 and public support beyond the first two fiscal years remains subject to review.
Korrun's 40 million yuan investment implies a 350.88 billion yuan value for DeepSeek through a fund and holding platform, but other disclosed investment routes, undisclosed share rights and reports of another financing make it a provisional transaction marker.
OpenRouter now lists Meta’s Muse Spark 1.1 at the same aggressive token price as Meta’s new developer API, adding interface compatibility without adding another inference provider. Early cache and reliability data make the offer easier to assess, but not yet proven for sustained agent workloads.
Nvidia's open-weight 8B Nemotron 3 embedding model is No. 1 on the RTEB beta leaderboard, but its published score covers 16 public tasks, the live table has no private-results column or visible mean for the leaders, and the smaller production model ranks 14th.
China has registered Apple Intelligence as a mobile on-device generative AI service, meeting a prerequisite to public release. Alibaba and Baidu have confirmed work with Apple, but Apple has not announced a launch date, assigned their exact roles or disclosed the partnership economics.
A rogue browser extension can trigger Claude for Chrome’s fixed Gmail, Docs, Calendar and business workflows; the attack is constrained by default approvals but can run silently for users who enabled unattended action.
Alibaba and ByteDance removed custom-agent features as China’s new companion-AI regime took effect, but the final rule exempts non-emotional assistants and still permits regulated adult companion services—raising the possibility that risk will move to specialist apps rather than disappear.
OpenAI says a handful of GPT-5.6 Sol file-deletion reports most commonly involved Full Access without sandboxing or Auto-review. Its own evaluations show a more complicated risk picture, and the company has not yet published an incident rate or evidence that its promised safeguards stop the failure.
A preprint credits GPT-5.6 Pro with finding a certified counterexample to a long-standing conjecture about the Benjamini-Hochberg procedure, but the proof concerns an asymptotic correlated Gaussian construction and the model-performance story rests on one reported research episode.
Twenty-nine countries signed an agreement creating WAICO as an independent intergovernmental organization, while China paired the launch with capacity-building offers that are not yet confirmed as WAICO programs.
Sunday Robotics says Memo folded and stacked garments successfully in 778 of 785 attempts across unseen homes without per-home adaptation, but the internal test covers one bounded chore and leaves price, support needs and reliability beyond laundry unanswered.
Beijing has discussed restricting overseas access to advanced Chinese AI models, while Washington is building a voluntary early-access review for cyber-capable systems. The approaches are not equivalent, neither implementing framework is complete, and downloadable weights make the moment before release the main point of control.
Gold Eagle has begun collecting and prioritizing AI-discovered software vulnerabilities, but the voluntary federal clearinghouse has not disclosed results, operating rules or the resources that would turn findings into deployed fixes.
OpenAI says its internal attacker helped harden GPT-5.6 Sol against prompt injection, but the benchmark scopes differ, humans still find attacks it misses, and neither the model nor enough cost detail is available for an independent comparison.
Dave Treadwell will replace departing AWS veteran Dave Brown as head of Compute and Machine Learning Services on August 1, moving one of the cloud unit’s largest internal customers into a supplier role as Amazon sharply increases infrastructure investment.
OpenAI and Work Louder’s limited-run Codex Micro puts six agent threads, common commands and reasoning settings on a physical controller, but the $230 device adds convenience rather than new Codex capabilities—and its undisclosed production run makes demand hard to judge.
Apple has reportedly sounded out chip startups and bankers after its Baltra server processor slipped, but the exploratory outreach has produced no disclosed target, price or deployment plan and sits alongside in-house chips, Broadcom custom-silicon agreements and Nvidia-powered Google Cloud capacity.
Nebius will let infrastructure partners finance, own and operate AI data centers connected to its cloud, while Nebius retains the software, service commitment and global sales channel. The model could add capacity with less company capital, but it supplements Nebius's own expansion and leaves partner economics and execution untested in public.