Mistral Raises €3 Billion in Samsung-Led Round at Valuation Above €21 Billion

01Mistral raises €3 billion in Samsung-led round at valuation above €21 billion

Mistral has raised €3 billion in a Series D funding round led by Samsung Electronics, giving the three-year-old AI company a post-money valuation of more than €21 billion. The company said the deal is the largest equity fundraising ever completed by a European technology company.

The financing gives Mistral more capital to pursue “sovereign AI,” its pitch to businesses and governments that want powerful AI systems without surrendering control of their data, models, computing infrastructure or production systems. Mistral says its open-weight models and private deployment options can also help customers avoid dependence on a single supplier’s roadmap, pricing or service availability.

Samsung led the round, with the EQT-managed Scaleup Europe Fund and existing investor PSG Equity serving as co-leads. New investors included Advent, funds and accounts managed by BlackRock, and the Grand Duchy of Luxembourg.

A large group of existing backers also participated, including a16z, ASML, Belfius, BNP Paribas CIB, Bpifrance, Carmignac, DST Global, Eurazeo, General Catalyst, Headline, Hillspire, Index Ventures, Korelya Capital, Lightspeed, Nvidia, Phoenix Court’s Solar fund and Salesforce Ventures. Mistral did not disclose how much each investor contributed.

The company said it will use the money to expand frontier research and the computing capacity needed to train powerful models. It also plans to build out infrastructure, accelerate commercial growth and extend its international footprint. Mistral did not specify how it will divide the funding among those areas or provide a timetable for the investment.

Mistral defines sovereignty across four parts of an AI system: data must remain within an organization’s boundaries; models must be controllable and customizable; computing resources must be private and predictable; and deployed production systems must be controllable and auditable. The company claims it is the world’s only AI provider building the full stack needed to offer that combination, including open-weight models, infrastructure, compute capacity and production products, though that claim was not independently verified in the announcement.

Mistral says it now operates in 20 countries and supports mission-critical AI work at more than 125 global enterprises, including Airbus, ASML and HSBC. The announcement did not disclose Mistral’s current revenue, cash consumption or the expected cost of its planned expansion.

02OpenAI Releases AI-Generated Navier–Stokes Proof as Mathematicians Question Research Leads and Credit

OpenAI has published what it says is an AI-generated proof of the Navier–Stokes existence and smoothness problem, one of seven Millennium Prize problems carrying a $1 million award for the first recognized solution. The equations are widely used to describe fluids such as water and air, but their theoretical behavior remains incompletely understood.

The company said an unreleased next-generation model produced the proof and that the final result was formalized in Lean, a programming language used to verify mathematical proofs. OpenAI began training a model with advanced mathematical capabilities on August 28 and intensified its effort after hearing that related research was progressing. More than 1,000 agents worked for over 50 hours, rising to as many as 10,000. OpenAI said the computing cost ran into millions of dollars; the week-long effort generated 300 billion output tokens, valued at $22.5 million at current Astra rates.

The announcement followed work by New York University mathematician Tristan Buckmaster and Anthropic researcher Levent Alpöge, who was not conducting the project for Anthropic. The pair published three proofs and a preliminary finding relevant to the Navier–Stokes problem, developed using models including Codex and Claude.

Buckmaster said information about their progress reached OpenAI before publication and that the company then concentrated substantial computing resources on a similar, rarely pursued route. OpenAI confirmed that its latest push began after rumors of progress on Millennium Prize problems, but said its researchers and agents did not see the pair’s prompts or proofs before their public release.

Because Buckmaster and Alpöge used Codex extensively, Buckmaster questioned whether their interactions could have influenced OpenAI’s model. OpenAI denied accessing specific user data to direct the project, while acknowledging that it could not entirely rule out de-identified product-use data having helped improve its models. No evidence establishes that the pair’s Codex data actually affected the result.

Buckmaster also alleged that OpenAI proposed announcing the solution without Alpöge’s name. OpenAI researcher Sebastien Bubeck denied suggesting that Alpöge’s credit be removed and recognized the pair’s priority on their separate unforced Euler work, which OpenAI said differs from its own results.

OpenAI’s Navier–Stokes proof still requires independent mathematical scrutiny, and the sources do not establish that it has passed peer review or met the Millennium Prize standard. The extent of human input, the complete record of the credit discussions, and any influence from Codex usage data also remain unknown.

03Anthropic Confirms Stolen Claude Sessions Are Consuming Subscription Allowances, but Users Cannot View Itemized Usage

Grant De Swardt knew something was wrong when his Claude allowance kept shrinking while he was not working. On August 5, the independent AI consultant paused or completed scheduled Cowork tasks, disabled cloud execution and confirmed that no local Claude Code task was active. His usage still rose from 45% to 55%.

De Swardt pays $200 a month for Claude Max 20x and relies on AI agents to run an automation consultancy serving small and midsize businesses. The unexplained consumption therefore threatened both his subscription allowance and his work. Yet when he asked Anthropic for an itemized usage record, the company did not provide one: support could see total consumption, but users could not inspect which tasks, sessions or services had used it.

Anthropic’s investigation found that a compromised Claude session key had been used to create unauthorized Claude Code OAuth tokens—credentials that let software access Claude Code without repeatedly entering a password. The company told De Swardt that an apparently unauthorized third-party service had used his account to process activity for other people. It could not determine how that service obtained access, saying the evidence was consistent with either stolen credential or session data, or a connection to an outside service.

Other subscribers reported similarly rapid consumption while they were idle, including one account that went from 0% to 49% in 12 minutes and another that exhausted its allowance daily for three days. Two users shared emails in which Anthropic said a malicious actor was using common “infostealer” malware—software that steals passwords, login credentials and saved session data—to capture Claude sessions and consume account usage. Anthropic said the malware did not originate from Claude itself.

The reports do not establish that every unexplained usage incident had the same cause. The source also gives no total number of affected accounts, duration of the activity or aggregate allowance consumed, leaving it unclear whether the campaign is continuing.

Anthropic suspended De Swardt’s paid account, invalidated his sessions and server-side Claude Code tokens, and refunded £44.49 for the remaining subscription period. It said that when suspicious activity is detected, users are signed out, existing authorizations are revoked, some refunds are issued and affected customers are advised to check for malware. De Swardt’s account returned after about two weeks, but he canceled his subscription. Anthropic declined to explain how users should identify misuse or whether itemized records would be introduced.