Samsung Leads Mistral’s Record €3B Round; US Flags China AI Distillation

AI & Machine Learning

Mistral AI raised €3 billion ($3.6 billion) in a Series D round led by Samsung Electronics on 8 September, lifting the French developer’s valuation to €21 billion ($24.4 billion) and making it the largest equity fundraising ever completed by a European technology company. EQT’s Scaleup Europe Fund and PSG Equity co-led, with participation from a16z, Nvidia, BlackRock, Advent, Salesforce Ventures and the government of Luxembourg. Mistral said the capital will fund compute capacity, infrastructure and international expansion, and reiterated a target of operating one gigawatt of European compute by 2030. Chief executive Arthur Mensch pitched the round around “sovereign AI,” giving customers regional control over where their queries run, while President Emmanuel Macron cast the Samsung tie-up as France and South Korea “building a third way” between the US and China. The deal shows how national-champion positioning has itself become a fundraising asset for non-US model builders. Source: TechCrunch Verified: True

Google DeepMind released AlphaGenome Atlas on 8 September, a roughly one-petabyte database of predicted molecular effects for all nine billion possible single-letter variants in the human genome. The resource precomputes outputs from DeepMind’s AlphaGenome sequence-to-function model across the whole genome rather than running the model one variant at a time, and the lab says it is more than 30 times larger than the AlphaFold database it expanded in 2022. It introduces the AlphaGenome Variant Impact (AVI) score, a single number that blends AlphaGenome and AlphaMissense outputs to rank a variant’s likely disruption to gene regulation or RNA splicing. In early testing the tool linked 19 genomic regions to body mass index and widened known non-coding associations for that trait by 22 percent. The Atlas is free for non-commercial research through a no-code web portal, with paid access on Google Cloud promised later. DeepMind stressed that it predicts molecular effects only and is not validated for clinical diagnosis. Source: Google Verified: True

Consumer Hardware

Dyson used a “Dyson Unveiled” showcase alongside IFA 2026 in Berlin on 3 September to launch a new robot-vacuum family called Nurovi, spanning three tiers. The flagship R3 uses infrared and green-laser illumination to spot invisible dried stains before scrubbing them, while the R2 Wash+Dry swaps the usual circular mop pads for an off-centre, curved triangular pad meant to reach into corners. Dyson also introduced the HushJet Big+Quiet air purifier, engineered with reshaped nozzles to keep airflow high while running quieter, plus a PencilWash W1 stick mop and the previously teased CameraJet toothbrush. The company said its 2026 theme is adding “more intelligence” — computer vision and machine learning — across the range. Most of the new machines are set for staggered release through late 2026. Dyson is not an official IFA exhibitor but stages its own event nearby each year to ride the show’s attention. Source: Engadget Verified: True

Huawei launched the Watch GT 7 Pro globally at a “Chase the Wild” event in Munich on 2 September, alongside the standard Watch GT 7, Watch 6 Pro, a MatePad Pro 12 tablet and nova 16s phones. The Pro pairs a titanium case and sapphire crystal with a 1.47-inch, 3,000-nit AMOLED display and 5ATM water resistance, and claims up to 21 days of battery life in light use. Huawei priced it at about €379 (roughly £380), with sales starting the same day through its own store and third-party retailers. New skiing, cycling and running modes plus a mini-workout assistant target outdoor users. Huawei continues to lean on wearables and tablets in Western markets, where US sanctions still restrict its access to Google Mobile Services and leading-edge chips. The GT line remains one of the few Huawei hardware categories with broad international distribution. Source: Huawei Central Verified: True

Cybersecurity

Microsoft’s September Patch Tuesday, released on 8 September, fixed a record 974 vulnerabilities across its products, including two flaws already exploited in the wild. CVE-2026-85880 is a heap buffer-overflow in the Windows Advanced Local Procedure Call (ALPC) subsystem that lets a low-privileged process escape its AppContainer sandbox and gain SYSTEM rights, while CVE-2026-81963 is an improper-link-resolution bug in the Windows Update stack that also yields local privilege escalation. Researchers noted the ALPC issue is only the second zero-day patched in that component in nearly four years, and the Update-stack flaw is the first ever flagged there. The release also addressed around 20 unauthenticated, potentially wormable remote-code-execution bugs. CISA added both exploited CVEs to its Known Exploited Vulnerabilities catalog the same day. The unusually large batch reflects both wider product coverage and a rising rate of vulnerability disclosure. Source: SecurityWeek Verified: True

Adobe rushed out an emergency hotfix on 7 September for CVE-2026-75650, a maximum-severity (CVSS 10.0) unauthenticated remote-code-execution flaw in Adobe Commerce and Magento Open Source dubbed “StyleSmuggler.” The bug abuses Magento’s template engine through PHP code injection triggered via the platform’s “Payment Transaction Failed Reminder” email, and Dutch security firm Sansec traced active exploitation back to 4 September. Attackers have used it to plant a Rust-based Linux backdoor and PHP web shells on storefronts, in some cases on sites that had applied every prior patch. Adobe’s hotfix VULN-39341 covers versions 2.4.4 through 2.4.9, and administrators are told to rotate every secret from admin passwords to payment-gateway API keys. CISA added the flaw to its KEV catalog on 8 September with an 11 September deadline for federal agencies. E-commerce platforms stay a favourite target because one flaw exposes payment data across thousands of merchants at once. Source: The Hacker News Verified: True

The NSA, CISA and FBI issued a joint cybersecurity advisory (AA26-251A) on 8 September accusing six China-based AI firms — DeepSeek, Moonshot AI, Alibaba, MiniMax, StepFun and Z.AI — of running “industrial-scale” knowledge-distillation campaigns to extract capabilities from US frontier models since at least late 2024. The agencies said the activity involved billions of tokens and millions of requests aimed at models from Anthropic, OpenAI, Google and xAI, and called distillation “the critical core” of the Chinese companies’ development rather than a supplement to it. To evade detection, operators allegedly spread traffic across native APIs, cloud resellers, pooled accounts and gray-market “transfer station” proxies that mask origin and bypass geographic limits. The advisory urges US labs to watch subscription-to-usage ratios, sudden maximum usage from new accounts and enterprise-scale throughput patterns. It stops short of announcing specific enforcement actions. The framing signals that model weights and reasoning traces are now treated as protected national-security assets. Source: CISA Verified: True

Enterprise Infrastructure

Qualcomm disclosed on 8 September that Amazon could buy up to $60 billion of its data-centre chips, networking hardware and related manufacturing services through 2036 under a new multi-generational partnership. The two will co-develop custom accelerators for AI inference — the workload of running already-trained models — and 1.6-terabit-per-second optical interconnects for hyperscale AWS clusters. Qualcomm granted Amazon warrants to buy roughly $4 billion of Qualcomm stock at $161.26 a share, 15 percent of which vested immediately and the rest tied to purchase milestones. Qualcomm’s finance chief said revenue should begin in the December quarter and become a “core component” of a $15 billion data-centre revenue target for fiscal 2029. Qualcomm shares jumped as much as 9.5 percent, their best day in years. The agreement lands Qualcomm its first major Western hyperscaler customer as it pushes beyond smartphone modems. Source: Mobile World Live Verified: True

Verizon and Corning announced a multi-year, multi-billion-dollar supply agreement on 8 September under which Verizon will buy more than 80 million miles of high-density optical fibre and connectivity products from 2027 through 2032. Verizon called the fibre “the physical engine for the AI era,” using it both to expand consumer and business broadband and to build a low-latency long-haul backbone linking data centres for AI hyperscalers. The deal underpins Verizon’s target of reaching 40 to 50 million broadband passings while positioning it as an infrastructure supplier to cloud providers such as AWS. Corning shares rose about 8 percent on the news, which landed the same day as Qualcomm’s Amazon deal and helped drive a broad rally in AI-infrastructure stocks. Corning’s Mike O’Day said generative AI and broadband expansion are “part of the same transformative opportunity.” Fibre supply has become a bottleneck as hyperscalers race to wire new data-centre campuses. Source: GlobeNewswire Verified: True

Policy & Regulation

China’s Ministry of Industry and Information Technology published a five-year plan on 8 September targeting 9,800 exaflops of “intelligent computing” capacity nationwide by 2030, more than four times the 2,185 exaflops reached by June 2026. The blueprint calls for 3.8 trillion yuan ($532 billion) of cumulative information-infrastructure investment over 2026 to 2030 and the “orderly” deployment of GPU clusters of 10,000 or even 100,000-plus accelerator cards. It repeatedly emphasises adapting data centres to domestic chips, reflecting Beijing’s drive to cut reliance on Nvidia and other US suppliers amid export controls. The plan extends the 2022 “East Data, West Computing” strategy of siting compute in western provinces with cheaper power. It arrives days ahead of a planned US–China dialogue on AI. The scale confirms that state-directed compute buildout is now central to China’s industrial policy. Source: South China Morning Post Verified: True

California’s Democratic-controlled legislature ended its session on 1 September having passed 26 bills targeting AI and social media, sending them to Governor Gavin Newsom, who has until the end of September to sign or veto. The measures would ban “addictive” feeds and autoplay video for users under 16, require age verification, bar employers from using AI to read workers’ “neural data” or emotional states, hold AI therapy chatbots to medical-confidentiality rules, mandate human instructors at California State University, and curb algorithmic “surveillance pricing.” OpenAI chief executive Sam Altman personally contacted Newsom to raise concerns about some provisions, according to Politico. The bills build on SB 53, the frontier-AI transparency law Newsom signed in 2025, further cementing California as the de facto US AI regulator in the absence of federal rules. Industry groups are lobbying hard for vetoes. Newsom’s choices will shape compliance obligations for platforms well beyond the state. Source: NBC News Verified: True