Akamai Lands an $11.6B Anthropic Deal as AI Chiefs Warn the UN

AI & Machine Learning

Frontier AI models have cracked two World War II Enigma messages that had resisted cryptanalysts for decades. Developer Carter Leffen used OpenAI’s Astra (GPT-6) to break a message that had stood unsolved since 2005, while cybersecurity executive Jack Willis used Anthropic’s Claude Opus 5 to decipher a separate one, with retired engineer and Enigma specialist Frode Weierud validating both plaintexts. Astra reportedly did its own archival research, pieced together context clues, built an Enigma simulator and recovered the text in about two days, work Weierud said would take a human researcher weeks or months. Only seven unsolved Enigma messages are now thought to remain, and the result is another marker of how far autonomous, multi-step research by AI agents has come. Source: TechCrunch Verified: True

OpenAI has set up an independent Advisory Group on Mathematics and Artificial Intelligence, hosted at Princeton’s Institute for Advanced Study, as its internal model racks up solutions to more than 100 open mathematics problems. The move follows an open letter from 25 Fields Medal winners criticising AI labs for racing each other to claim famous problems. The nine initial members will judge the significance of results and help coordinate how they are released, but they will have no authority to slow or redirect OpenAI’s research. Members are unpaid and free to speak publicly, and only one of them, Camillo De Lellis, also signed the Fields Medallists’ letter. Source: TechCrunch Verified: True

Consumer Hardware

Meta’s Muse AI agent app is surging after Meta Connect, with downloads climbing from 2.5 million at the start of the week to 3.4 million by Thursday according to Sensor Tower, and daily active users jumping 27% the day after the keynote. Muse has held the top spot on the US App Store since September 18 and on Google Play since September 19, and its early download growth is running at roughly double ChatGPT’s launch pace. Meta is backing the app with heavy house advertising across Facebook, Instagram and WhatsApp plus Reddit, TikTok and YouTube, though ads account for only a small share of impressions so far. At Connect it previewed avatar video chat, Mac computer use, a dedicated Muse email address, more connectors and smart-glasses integration, and it has now opened an early-access programme for those features. Source: TechCrunch Verified: True

Startup Meadow has unveiled a small, square Android phone designed as a companion to a main smartphone rather than a replacement, aimed at people who want to leave their main device at home. The device has a 3-inch touchscreen and runs essential apps such as Maps, Strava, Uber, Spotify and Apple Music, alongside built-in calling, texting, notes, voice memos, location sharing and ticket scanning. A small mirror next to the camera lens handles selfies, and the phone uses its own number through a Meadow subscription plan. Founded by 25-year-old Shreyas Narlanka, whose previous startup was acquired by StockX, Meadow is backed by basecase capital and angels including Vercel founder Guillermo Rauch. Source: TechCrunch Verified: True

Cybersecurity

Kiteworks, the file-transfer company formerly known as Accellion, has urged customers to shut down their servers before the weekend after law enforcement warned of a credible, imminent cyberattack that may involve zero-day flaws the company has not yet identified. CISO Frank Balonis said the threat intelligence indicated an actor “may attempt to target some Kiteworks systems,” and the company is also pushing customers to upgrade to version 9.5.1. Kiteworks serves thousands of organisations in healthcare, education, automotive, technology and government, and at least one healthcare customer reported disruption to patient contact. The warning carries extra weight because Accellion’s file-transfer appliance was at the centre of a 2020 breach campaign that compromised hundreds of organisations. Source: TechCrunch Verified: True

A pre-authentication SQL injection flaw in Roundcube Webmail, tracked as CVE-2026-48842 with a CVSS score of 8.1, is now being actively exploited according to Canada’s Cyber Centre. The bug lives in the virtuser_query plugin and stems from a preg_replace() backslash-escape bypass that lets unauthenticated attackers inject SQL into the database backend, potentially exposing mail credentials and stored messages. Fixes shipped in May in versions 1.6.16 and 1.7.1, but Shadowserver counts more than 523,000 Roundcube instances exposed to the internet. Roundcube is a repeat target, with two earlier flaws added to CISA’s exploited-vulnerabilities catalogue in February. Source: The Hacker News Verified: True

Revolut customers have been hit by a second data breach this month after attackers socially engineered their way into DriveWealth, the US brokerage that once held Revolut users’ stock-trading accounts. The intrusion on September 4-5 exposed names, email and postal addresses, phone numbers, employment details, citizenship, age, gender and partial account numbers, though passwords and payment data were not taken. The data is historical, because Revolut moved customers off DriveWealth between December 2023 and June 2025, and neither company has said how many people are affected. The incident is a reminder that third-party and legacy partners remain a soft spot in fintech supply chains. Source: The Register Verified: True

Enterprise Infrastructure

Anthropic has agreed to pay Akamai $11.6 billion over seven years for cloud computing capacity, the largest deal in Akamai’s history, with room to expand to roughly $20 billion. The contract is focused on CPU capacity rather than GPUs, a sign that agentic AI workloads are driving demand for general-purpose compute as well as accelerators. Akamai expects $150-300 million of revenue from the deal in 2027, rising to about $1.7 billion a year by the end of 2028, and Anthropic receives warrants for up to 5% of Akamai’s shares at $111.33 each, vesting as its spending grows. Akamai’s stock rose as much as 17% on the news, which extends Anthropic’s run of compute deals across multiple providers. Source: TechCrunch Verified: True

British AI neocloud Nscale has secured $3.36 billion in convertible financing ahead of a planned New York Stock Exchange listing that could value it at around $35 billion. Hedge fund Third Point led the round, with $2.36 billion available immediately and a further $1 billion from existing investor Nvidia due in mid-November, and the notes will convert to equity when the IPO completes. Nscale, spun out of Australian crypto miner Arkon Energy two years ago, says it has signed more than $103 billion in contracts and is building large data centre campuses in Norway and West Virginia. The raise shows how much capital AI infrastructure specialists are pulling in as they race to add capacity. Source: TechCrunch Verified: True

AI data centre builder Crusoe has cancelled a $1.25 billion agreement to buy 29 of Boom Supersonic’s 42-megawatt Superpower gas turbines, which had been due to start arriving in 2027. Crusoe said the turbines “aren’t the right fit today” and that it will instead lean on wind, solar, batteries and grid power for its sites. Boom launched the turbine business last year on the back of its Overture jet engine work, with about 80% shared components, and raised $300 million partly to fund it, so losing its anchor customer is a significant blow. The reversal highlights how quickly power strategies are shifting for AI campuses as operators balance speed, cost and emissions. Source: TechCrunch Verified: True

Policy & Regulation

The UN Security Council held its first high-level briefing dedicated to the safety risks of advanced AI, hearing from OpenAI CEO Sam Altman, Anthropic CEO Dario Amodei, Hugging Face co-founder Clément Delangue and Turing Award winner Yoshua Bengio. Amodei told the council that “if managed poorly,” AI could be a risk to humanity as a whole, while Altman argued the most important decisions cannot be left to labs in San Francisco alone, and Bengio called the dangers “real and imminent.” Speakers pressed for international cooperation, including US-China work on a notification system for AI-related national security incidents. The session came a day after President Trump used his UN General Assembly speech to reject global AI governance frameworks, underlining the gap between the labs’ calls for coordination and Washington’s position. Source: BNN Bloomberg Verified: True

Weeks after safety talks between rival labs came to light, Google, OpenAI and Anthropic are now moving to set up a formal self-regulatory body, the Standards Authority for Frontier AI (SAFA), which could launch by the end of 2026 or early 2027. Modelled on the financial industry’s FINRA, the group would develop common benchmarks and run safety testing on advanced models. The labs have approached former White House AI adviser Sriram Krishnan about the CEO role, and have also sounded out Arati Prabhakar, Condoleezza Rice and David Friedberg, with METR’s Beth Barnes and Paul Christiano as possible scientific advisers. As an industry body SAFA would not need congressional approval, though it may need government registration to carry legal weight, and critics are likely to question whether labs should write their own rules. Source: BankInfoSecurity Verified: True