Microsoft cloud outage takes down Outlook, Teams and Copilot worldwide

Microsoft said on Friday that its cloud services were recovering from a worldwide outage. The failure began on Thursday in the Americas and spread across regions through the day. Outlook, Teams, SharePoint, Copilot, Defender and the Intune device-management tools were all affected. Some customers lost webmail, files and video meetings for much of two days. Microsoft said others still faced intermittent access and delayed mail as it worked through backlogs.

The company said it had found no evidence of malicious activity. It will publish a full root-cause report once mitigation is complete, and has given no timetable for full restoration. Friday updates reported broader improvement in North America, sustained gains across much of Europe and recovery signals in Asia Pacific. Network monitoring firm Kentik said traffic to Microsoft’s services was returning after about 18 hours of disruption. Microsoft’s previous major incident, a days-long outage in September, was blamed on a core authentication configuration error.

Featured source: Microsoft service health , The Verge , TechCrunch .

Other Stories

Democrats open a second Nvidia inquiry as Huang agrees to testify

Representative Richard Neal and Senator Ron Wyden asked Commerce Secretary Howard Lutnick to preserve records on export licences for Nvidia’s H200 sales to China. The two are the senior Democrats on the House Ways and Means and Senate Finance committees. Their letter asks why Megaspeed, a Singapore-based Nvidia customer, kept buying chips after US intelligence raised concerns. It also asks why other Asian buyers with no physical presence at their listed addresses were approved. Democrats in the minority can request documents but cannot compel their release. Jensen Huang, removed as chief executive on Thursday, has separately agreed to appear before Republican-led committees that had threatened a subpoena.

Filed from: TechCrunch .

Google puts Gemini 4 Hydrogen into Search worldwide

Google said Gemini 4 Hydrogen is now the model behind Google Search for all users worldwide, calling it the largest model upgrade to Search so far. The company says it handles multi-part questions, runs several queries at once and returns tables, images and links. Hydrogen costs $1.50 per million input tokens and $7.50 per million output tokens, with a one-million-token context window at a flat price. Google reports a score of 51 on the Artificial Analysis Intelligence Index at the default thinking level, against 50 for its larger Argon model. It reports 54 at the model’s highest setting. Those figures come from Google and the benchmark’s operator.

Filed from: Google , Google Search .

Nebius plans $40 billion of 2027 spending as a Microsoft site slips

Nebius told investors on its third-quarter call that it plans about $40 billion of spending in 2027, more than analysts expected. It intends to fund the buildout with customer prepayments and loans backed by customer contracts. Its contract with Microsoft is worth up to $19.4 billion. At a Vineland, New Jersey, site, only just over half of the planned capacity will arrive by the end of this year. The rest is due in the second quarter of 2027. The site’s operator wound down temporary generation faster than expected while switching to fuel cells. New Jersey regulators had fined it $1.1 million over unpermitted gas generators.

Filed from: TechCrunch .

European security buyers move spending toward local vendors

Sales from EU-based cybersecurity vendors grew 35% in the two years to the third quarter of 2026. The figure comes from a Context report first covered by The Register. The market overall grew 10% and US vendors grew 8% over the same period. US suppliers still hold 62% of the EMEA market. They also lead in identity and cloud security, a gap the report tentatively links to continued use of US cloud platforms. Buyers now ask who holds encryption keys, who can read sensitive logs and where data is stored, a Context research vice president said. The demand signal is clearer than the claims behind it. An October Omdia study found 28% of providers do not disclose evidence for their sovereignty claims, and no regulator polices the label.

Filed from: TechRadar Pro .

Brain-implant volunteers publish their own trial guidelines

Members of the BCI Pioneers Coalition published recommendations on Tuesday for researchers and companies running brain-computer interface trials. The group is made up of people who have taken part in those trials. Its guidance covers 13 categories, from paying participants for time that can amount to a part-time job to crediting them as collaborators or co-authors. It also addresses protection of neural and personal data and the wording of consent and nondisclosure terms. Ian Burkhart, who had a 96-electrode array implanted in 2014 and took part in two trials over seven and a half years, co-founded the group in 2022. The authors note that not every participant wants public credit.

Filed from: MIT Technology Review .

From the Community

Apple opens its Foundation Models framework to other providers’ models

Apple’s documentation now describes the Foundation Models framework as a single Swift API. It reaches the on-device Apple Intelligence model, Apple’s Private Cloud Compute and server models from other providers. Any provider can ship a Swift package conforming to a language model protocol, the page says. Guided generation, tool calling and image understanding then work across those paths. It also lists Dynamic Profiles for changing models mid-session, an evaluations framework and a Python SDK. Apple says developers in its Small Business Program can use its models on Private Cloud Compute without cloud API charges. Daily limits apply, and the company says they may later carry a fee. The page is documentation rather than a release announcement, and gives no ship date.

Filed from: Apple Developer .

Brave lets agents pay for search results per request

Brave added the open x402 payment protocol to its Search API, so developers and agents can pay per query instead of managing API keys. An endpoint behind x402 answers an unpaid request with HTTP 402 and a price. The agent signs a USDC payment on the Base blockchain, which Brave verifies before returning results with a receipt. Pricing follows Brave’s published rate of $0.005 per request, or $5 per 1,000. It applies to the agent-oriented LLM Context endpoint and to web, news, image and video search. Brave runs its own payment facilitator and is covering Base gas fees for a limited time.

Filed from: Brave .