Skip to content

AI Flash · live feed

What is happening in AI right now

A timeline of everything that has passed through the radar — newest first, with an evidence status for each item.

17 new news items today 1147 across the entire feed ↗ details, original sources and our own analyses

Continuously updated

Latest items

1147 items in the selected feed

How to read evidence statuses ↓

Wednesday, July 22

24 items
Regulation and law only one source so far

US administration divides over its response to advances in Chinese AI models

The White House and the US administration's Commerce Department disagree on the approach to Chinese AI models. The White House wants strict controls against distillation attacks (training models based on US systems), while the Commerce Department considers this impracticable. The reason is concern about the model…

Wired — AI section Original source ↗
AI agents only one source so far

Monday.com lays off 20 % of employees to focus on AI

Monday.com is laying off 630 employees (20% of its workforce) and restructuring operations around the AI Work Platform. The new strategy focuses on a no-code app builder, AI agents and workflow automation. The move is part of a broader trend — more than 122 000 tech roles have already been cut in 2026, while 78 % of companies…

TechCrunch AI Original source ↗
Other clearly official source

Claude Services outage resolved

Claude Services experienced an outage affecting claude.ai, Claude Code, Claude Code Web and Cowork Remote. The issue was identified and the incident is resolved.

Claude Status — incident history Original source ↗
Security only one source so far

Arcee says Chinese models are not inherently dangerous

Arcee CTO Lucas Atkins says Chinese open-weight models (Qwen, Kimi K3) are no more dangerous than other open-source software. It is technically impossible for their creator to retain control after they are run locally. Security risks are theoretically possible but very difficult in practice.

TechCrunch AI Original source ↗
Tools and apps confirmed by 2 independent sources

Substack adds AI content detection in newsletters

Substack launched a tool for detecting AI content in posts and newsletters through an integration with Pangram. It lets users see the estimated proportions of AI and human writing. Authors can add a note about their use of AI.

The Verge AI · and 1 more source Original source ↗
Tools and apps clearly official source

A new Copilot impact metrics dashboard

GitHub released a new dashboard for enterprise administrators to visualize Copilot adoption in organizations. It shows users in three adoption stages, with metrics including average merged pull requests per month, merge speed and lines of code per day. It helps identify inactive users and provides recommendations for…

GitHub Changelog (Copilot and AI features) Original source ↗
Business and investment only one source so far

OpenAI raises planned infrastructure spending to $750 billion by 2030

OpenAI announced a 25% increase in planned infrastructure spending to $750 billion by 2030. The first major project is the Project Camellia data center in Georgia (1400 acres), requiring 3.2 GW of electricity from Georgia Power, fully financed by OpenAI. Capacity will be available in 2028–2032. Regulatory filings suggest that the energy will come mainly from solar power, batteries and natural gas.

TechCrunch AI Original source ↗
AI agents only one source so far

How monday.com runs AI agents on Amazon Bedrock in production

monday.com runs Sphera, an internal agent system with Atlas, an agent that takes tickets, writes PRs and deploys features. Agents have stable identities, are part of teams and deliver a 50% increase in output per engineer. The architecture uses AWS services (Bedrock, SQS, EKS, RDS).

AWS Machine Learning Blog Original source ↗
Business and investment only one source so far

Microsoft and Mistral partnership to expand sovereign AI infrastructure in Europe

A partnership between Microsoft and French AI startup Mistral AI, announced on 21 July, will expand AI infrastructure in Europe. Mistral Medium 3.4 and OCR 4 will be available in Microsoft Foundry and Copilot Studio; Microsoft will use Mistral data centers with Nvidia Rubin GPUs. The partnership strengthens Mistral's position as a leading European AI vendor and gives European companies more sovereign solution options.

Business and investment only one source so far

European Commission launches three GenAI pilot projects for public administration

The EU is launching FLOODS & DROUGHTS, EUNOMIA.AI and EuropAI, projects funded by the Digital Europe Programme. They aim to help public administrations develop, procure and deploy generative AI solutions in line with European legal and safety standards. The kick-off is scheduled for 14 September 2026.

European Commission — Shaping Europe's digital future (AI Act, EU regulation) Original source ↗
Open-source only one source so far

A podcast on open-source AI model development: Kimi K3, Qwen and geopolitics

An Interconnects podcast in which Nathan Lambert and Florian Brand discuss open-source AI model development. Topics include Kimi K3, Qwen's announcement that its next model will be open, Xi Jinping's speech on openness as a strategy, benchmarking, distillation and the gap between open and closed models.

Interconnects (Nathan Lambert, AI2) Original source ↗
Regulation and law only one source so far

Verona covers Flock cameras after Flock refuses to remove them

Verona, Wisconsin, tried to remove three Flock automated license plate readers after deciding not to renew its contract. However, Flock did not remove the cameras despite repeated requests. The city was unsure whether it had legal authority to remove them itself, so it covered them with black…

404 Media (investigative tech journalism) Original source ↗
Security only one source so far

Podcast: Flock tracks people, not just cars

The podcast features stories about police surveillance of people using Flock, secret sales of books to AI companies for training and the hacking of Suno. Available on Apple Podcasts, Spotify and YouTube, with a bonus for subscribers.

404 Media (investigative tech journalism) Original source ↗
Other only one source so far

The US Army used up its annual AI token budget in a month

The US Army exhausted its annual token budget on the Ask Sage AI platform just a few weeks after announcing it as unlimited in May 2026. The service supports Gemini, Llama and ChatGPT models. It is unclear whether the budget will be renewed after 1 October.

Ars Technica (AI) Original source ↗
Robotics only one source so far

Agibot expands its embodied AI product portfolio with four new robots

At a conference in Shanghai, Chinese company Agibot introduced four new products: the A3 Ultra humanoid (5 kg payload per arm, up to 8 hours of operation), the X2 EDU research platform, the G2 Max industrial robot for handling heavier loads and the OmniHand 3 Ultra-M dexterous hand with 20 degrees of freedom. The aim is…

Business and investment only one source so far

Energy companies pledge not to pass AI costs on to consumers

Nearly 200 organizations, including NextEra Energy, Duke Energy, Equinix and Digital Realty, signed Trump's "rate payer protection pledge" to protect consumers from rising electricity prices caused by data centers. According to the White House, these companies distribute approximately 80 % of US electricity. The addendum is…