Archive
Every story we have published, newest first. Search by company, product, or idea, or filter by topic.
A reporting rule is the bigger move.
The cases include hidden instructions, unapproved file uploads, and messages between models. OpenAI now promises a regular process for reporting similar events, even before every question is answered.
The numbers still need outside tests.
The Atlas 960E SuperPoD links up to 4,096 AI processors with near-packaged optics. Huawei claims lower power use and higher reliability, while the first Ascend 960 chips are planned for 2027.
The bill is only a first step.
The House passed the Ratepayer Protection Act by 417 votes to 3. It asks state regulators to consider special rules for sites above 100 megawatts, but it does not set one national price.
A private summit is not public control.
Leaders from OpenAI, Anthropic, Google DeepMind, and Nvidia met ministers and civil-society figures in Scotland. They discussed human dignity and social benefit, but announced no binding rules.
The AI power boom now has a reliability bill.
Generac expects $2.4 billion of initial deliveries in 2027 and 2028. Amazon also received rights to buy shares as payments for data-centre generators rise toward a possible $8 billion.
A new alliance says flexible power use is the deal.
Google, Nvidia, Emerald AI, and energy companies launched a coalition for data centres that can reduce or move electricity use when the grid is stressed. The promises now need measured results.
It is still a research bet.
The two governments plan to fund Yoshua Bengio’s nonprofit LawZero. Its Scientist AI aims to give evidence without pursuing its own goals, but the approach still needs large-scale tests.
Recovery is the useful idea.
Agent Resilience can map an agent’s memory, settings, and connected systems, then restore a trusted state. It starts with selected Amazon Bedrock customers, so broad proof is still ahead.
“Secure” still needs a clear meaning.
Missionforce plans to use OpenAI models through Amazon Bedrock and connect some work to ChatGPT. New policy and operations tools promise speed, but public agencies need strict review and audit rules.
The test is whether research reaches real factories.
The EIT programme plans to support founders, fund scale-ups, and connect robotics teams with industry through 2028. Its targets are useful, but they are goals rather than delivered products.
Permissions are the real interface.
AIforce connects Salesforce records and workflows to Claude, Slack, Amazon tools, and Gemini Enterprise. The idea is useful, but each action still depends on access rules, review, and clear availability.
Its strongest results come from Salesforce tests.
Koa is based on Nvidia Nemotron and trained with synthetic business tasks. Salesforce says it makes fewer CRM action errors, but the benchmark and the customer pilots still need outside study.
Working safely beside people is the bigger test.
Agility says its new humanoid can lift 22.7 kilograms, work for 90 minutes, and charge in nine minutes. Early access starts in 2027, so its safety and productivity claims are not yet broad field results.
The speed cost is large.
Open-1b publishes checkpoints, data, code, and hashes for every training step. Gensyn says audits work across several kinds of hardware, while its reproducible runtime is about five times slower than an optimised system.
Owners still need a safe way to start small.
Anthropic added 43 workflows and 27 new integrations for leads, invoices, marketing, hiring, and operations. Some can run on a schedule, so owners must set narrow rights and review important messages or payments.
The beta starts with useful limits.
Apple began the English beta on September 14. Siri can use messages, email, photos, screen content, and app actions, but access depends on device, region, language, and daily cloud limits.
The hardware is still the hard part.
CUDA-Q Logical links quantum programs, error-correction methods, and hardware estimates. Fermilab reports faster design work, while Sandia’s QUOPS benchmark offers a common test for future systems.
It does not measure every job.
The open Atlas compares work-related AI use across countries and occupations. A linked science study reports time savings, but also finds that checking results and physical work still slow research.
The company updated its voter-information campaign for the 2026 US elections. It asks people to open citations, use official local election pages, and treat AI output as a starting point.
The adviser still owns the decision.
Anthropic released connectors and workflow skills for meeting preparation, portfolio review, records, and compliance checks. Client advice and other regulated work still require human approval.
Questions are not findings.
Two senators asked OpenAI for details about the July breach, its safety tests, and outside access to evidence. OpenAI says it investigated the incident, but the new Senate requests seek records that officials can check.
A letter is not a law.
The members asked Speaker Mike Johnson to keep the House in Washington until it advances bipartisan AI rules. They pointed to existing bills, but leaders had not promised a vote when the letter became public.
The plan now needs rules that rivals can verify.
Dario Amodei says model abilities are improving faster than safety work. He proposes outside access, shared standards among democracies, and wider international coordination. His timelines are warnings, not proven forecasts.
Its Rust rewrite is the smaller story.
OpenAI says Habitat handles more than 70 million requests each second and over 500 petabytes of data. A new Rust service uses less computing power, but reliable limits and simple data paths made the scale possible.
Production is still planned for 2027.
Liberty Global joined a funding round that values Positron at $5 billion. The company says its design avoids scarce high-bandwidth memory, but its next chip has not entered production.
The cases are warnings, not a count of all abuse.
Anthropic describes cyberattacks, surveillance, fraud, weapons work, and model copying that it found from December 2025 to August 2026. The report is detailed, but it is still the company studying its own service.
A new package requires stronger chatbot safeguards, bans some addictive feed features for users under 16, and adds child privacy rules. The next test is whether regulators can make the rules work in real products.
None of the new spacecraft is flying yet.
Tomorrow.io explained how its next satellite system would measure storms, wind, moisture, clouds, and rain. Better data could help AI forecasts, but the company has not published the final fleet size or real warning gains.
An investigation is not a finding of wrongdoing.
Reports say the Justice Department is studying a licensing deal that moved Groq leaders to Nvidia without buying the company. Nvidia’s accounts show $17 billion in payments and obligations tied to the agreement.
Most of the unified experience comes later.
The planned Enterprise AI Harness joins business data, actions, security, models, and governance. Some parts exist now, while the unified product is due to start arriving in early fiscal 2028.