HMRC Brings in Quantexa for £175m Data Cleanse Ahead of AWS Move
HMRC is investing £175 million (approximately $220 million) to clean and modernise its data ahead of a massive migration to Amazon Web Services by 2028. The tax authority has signed a ten‑year partnership with British data‑analytics firm Quantexa to overhaul its core data infrastructure, aiming for a unified view that supports AI and tax‑risk detection. This follows a separate £473 million ($633 million) AWS contract to move applications from three Fujitsu data centres to the cloud. The effort comes amid criticism that UK government data remains fragmented and poorly standardized.
UK Government's "Pre-Seed to Growth" Fund Invests in Google-Owned AI Lab's $2.1bn Series B
The UK government’s new Sovereign AI fund has taken a strategic stake in Isomorphic Labs, the DeepMind‑spun‑out that just closed a $2.1 billion Series B round. The fund targets “pre‑seed to growth” AI ventures, positioning Britain as a hub for early‑stage artificial‑intelligence...
Salesforce Tools up for 100K Big, Concurrent Chats with AI Data Distribution Layer
Salesforce unveiled a new AI data distribution layer that enables up to 100,000 concurrent chat sessions, tackling the performance bottlene‑bottles that arise at scale. The redesign addresses Kafka lag, uneven tenant distribution, and Postgres hotspot issues while preserving real‑time AI...
Coinbase Thought It Could Failover. It Didn't Count on Bug in Its Amazon MSK Deployment.
Coinbase experienced a roughly five‑hour outage after a bug in its Amazon Managed Streaming for Apache Kafka (MSK) deployment prevented a smooth failover. The incident began in an AWS US‑East‑1 data centre but quickly spread as the platform could not...
Pwn a CEO with a Single Email? Patch Tuesday Brings Nasty Zero-Click Outlook Bug
Microsoft’s May 2026 Patch Tuesday introduced a critical zero‑click remote code execution vulnerability affecting Outlook. The flaw, originally cataloged as a Word issue, actually enables attackers to execute arbitrary code simply by sending a crafted email, with no user interaction...
NHS to Spend £900 Million on Healthcare AI
Britain’s National Health Service announced a £900 million (≈$1.15 billion) investment in artificial intelligence for healthcare. The funding will be allocated over an eight‑year period beginning in 2027. The program aims to accelerate AI‑driven diagnostics, patient triage, and operational efficiencies across NHS...
Mini Shai Hulud Strikes Again Hitting over 100 Npm and PyPI Packages Including Mistral AI
The Mini Shai Hulud supply‑chain campaign has resurfaced, this time compromising more than 100 npm and PyPI packages, including popular TanStack libraries and Mistral AI’s TypeScript client. The attack spreads by harvesting CI/CD secrets and injects a credential‑stealing payload that...
Google Says LiteRT Is Almost There for Any-Device, Edge Agentic AI – and Beats the Pants Off Llama
Google has rebranded its TensorFlow Lite runtime as LiteRT and says the framework is now close to supporting agentic AI workflows on any edge device. The updated system promises model‑agnostic deployment, allowing developers to run popular open‑source models locally with...
Linux Distro Debian Goes All in on Reproducible Software
Debian, one of the oldest Linux distributions, announced a new security policy that mandates all future packages be reproducible, meaning they can be rebuilt byte‑for‑byte from source. The release team has updated its migration tooling to automatically reject any package...
STACKUP: The Stack's Weekly Tech Startups and Funding Wrap
The Stack’s weekly wrap highlights Europe’s quantum surge, with two startups each raising over $100 million, led by Dutch firm QuantWare’s $110 million Series B to boost processor production. A Chinese AI laboratory announced a new breakthrough model, adding pressure to the global...
OpenAI's Secret Sauce for Networking Is Now an Open Protocol. What Does It Mean for Inference?
OpenAI unveiled Multipath Reliable Connection (MRC), an open networking protocol designed to speed and stabilize data transfer between GPU clusters in massive supercomputers. The protocol, developed with AMD, Broadcom, Intel, Microsoft and NVIDIA, can link more than 130,000 GPUs using...
Data Centre Fire Knocks IBM's Cloud Service Offline
A fire broke out at NorthC's 11 MW data centre in Almere, Netherlands, on May 8, forcing a power shutdown that knocked IBM's Amsterdam 03 cloud region offline for roughly 12 hours. While the blaze was confined to the rear of the 26,000 m² facility...
AWS Outage Halts Coinbase Trades
AWS experienced a thermal event in its US‑EAST‑1 data centre on May 8, causing overheating of server racks and forcing EC2 instances, EBS volumes and other services offline. The outage prompted AWS to add cooling capacity and shift traffic away...
Is Fully Homomorphic Encryption - Cryptography's "Holy Grail" - Inching Closer to Mainstream Use?
Fully homomorphic encryption (FHE) allows computation on encrypted data without decryption. After decades of theoretical work, Intel unveiled its Heracles chip in March 2026, delivering roughly 5,000× speedup over its best CPUs for FHE workloads. The five‑year hardware program targets...
EU Kicks AI Act's Strictest Regs Down the Road
The European Parliament and Council have agreed to push back the EU AI Act’s high‑risk provisions from August 2026 to December 2027 and to defer rules for AI embedded in products, such as toys, to August 2028. Tech Commissioner Henna...
Arm Sees $2bn Demand for New CPU, Before Shipping a Single Chip
Arm announced that its upcoming AGI data‑center CPU has already attracted roughly $2 billion in pre‑orders, despite not having shipped a single chip. The company’s CEO Rene Haas said the new processor will make data‑center services Arm’s largest business segment. The...
AXA's Fraught Journey to the Cloud Starts to Pay Off
AXA, the French‑based global insurer, has finally seen tangible results from its decade‑long cloud migration effort. After a rocky virtual‑machine lift‑and‑shift, the company now runs heavily containerised applications in public clouds. The new architecture powers critical claim‑processing and underwriting workloads,...
Allianz Hands Cyber Insurance Off to San Francisco-Based Specialist 'Coalition'
Allianz has entered a ten‑year agreement with San Francisco‑based cyber insurer Coalition, handing over full management of its commercial cyber insurance portfolio. Coalition will lead pricing, product development, risk mitigation and claims, while Allianz retains branding, distribution power and takes an...
The AI Tool Google Says Can Speed up LLM Inference by 3x
Google’s DeepMind unveiled multi‑token prediction drafters for its open‑source Gemma 4 models, using speculative decoding to boost inference throughput. The technique can deliver up to three times faster generation without degrading output quality or reasoning. By cutting compute per token, it...
Aon Touts 95% Productivity Gains From AI Tools, but Says It Won't Cut Headcount
Aon announced that its suite of AI tools has delivered a 95% boost in productivity across the firm, from underwriting to client analytics. Despite the dramatic efficiency gains, the company’s CEO stressed that Aon does not plan to reduce headcount,...
DARPA Wants Robots that Compute with Their Bodies' Material
DARPA has issued a request for information on a program called “Rethinking robotics with physical intelligence,” seeking materials that embed sensing, computation and actuation directly into a robot’s structure. The aim is to collapse processing loops so robots can react...
Mythos May Be Pushing the US Toward Stricter Regulation It Said Would Kill AI
Mythos, a leading AI developer, is lobbying for U.S. legislation that would require government review of powerful AI models before they are released. The proposal mirrors earlier calls for “pre‑deployment” oversight, but Mythos argues the rules would be so restrictive...
Coinbase Slashes Headcount, Sees "Fleets of Agents" Instead
Coinbase announced it will lay off roughly 14% of its staff—over 690 employees—as it pivots toward AI‑driven development. CEO Brian Armstrong said AI now lets engineers ship weeks‑long projects in days, prompting a shift to "AI‑native" talent that can manage...
Copy Fail Exploitation Has Begun, and Brian Pak Is Sorry for the Chaos
The U.S. Cybersecurity and Infrastructure Security Agency (CISA) has placed CVE‑2026‑31431, known as "Copy Fail," on its Known Exploited Vulnerabilities (KEV) catalog, confirming active attacks against Linux systems. The flaw, disclosed only days earlier, enables privilege escalation by exploiting a...
STACKUP: The Stack's Weekly Tech Startups and Funding Wrap
Online Oceans, a UK maritime‑tech startup, closed a £4 million ($5.5 million) funding round led by Seraphim Space, with angel investors including SolarCity co‑founder Peter Rive. The company builds solar‑powered autonomous surface vessels, called Scouts, to collect data for maritime and defence...
The Big Interview: Johanna Hutchinson, CDO, BAE Systems
Johanna Hutchinson, BAE Systems' first chief data officer, joined the defense giant in 2022 to turn data into a strategic asset. BAE, with 111,000 staff and an order backlog of £83.6 billion (about $106 billion), relies on Hutchinson to streamline delivery of...
Home Office Eyes £296m Refresh for Biometrics Platform
The UK Home Office has announced a £296 million (approximately $376 million) investment to refresh its national biometric platform, the Strategic Central and Bureau Platform (SCBP). The funding will support continued modernization after the system’s recent migration to modern technology stacks. SCBP...
Ubuntu Maintainer Canonical: We’re Under Attack
Canonical, the company behind Ubuntu, confirmed that its web infrastructure has been hit by a sustained, cross‑border DDoS attack, leaving key pages—including the security advisory site—unavailable for more than 14 hours. The group calling itself "Islamic Cyber Resistance in Iraq...
Improving AI Accuracy with GraphRAG
AWS’s managed graph database Amazon Neptune is gaining traction as a catalyst for higher AI accuracy, especially in security and chatbot applications. Customers such as Trend Micro have lifted chatbot precision from 70% to 90% by leveraging Neptune’s relationship‑focused data...
Linux Bug “Copy Fail”: Short Python Script Gives Root on… Everything?
A newly disclosed Linux kernel vulnerability, dubbed “Copy Fail,” allows a short Python script to gain root privileges on any distribution. The exploit was crafted after Xint Code fed a large language model the kernel source for roughly an hour,...
IPv8: Lone Engineer Proposes Simpler Internet Addressing, Triggers Outcry
Network veteran James Thain, seeking a turnkey server stack, drafted a new Internet Protocol dubbed IPv8 in just five days with AI assistance. He submitted the proposal to the IETF, where it was logged as an individual contribution. The submission...
GitHub CTO: We’re Going Multicloud Amid AI Stress
GitHub’s chief technology officer announced that the code‑hosting platform is shifting to a multicloud strategy to handle the surge in AI‑driven workloads. The move will spread GitHub’s services across Azure, AWS, and Google Cloud, reducing reliance on any single provider....
What's New About Cisco's Sovereign Infrastructure Push?
Cisco has launched a suite of networking solutions specifically engineered for sovereign critical infrastructure across the entire EMEA region. The offering, branded as "sovereign‑ready," bundles hardware, software and security services that comply with local data‑sovereignty regulations. Cisco positions the portfolio...
UK Gov't Teases New "AI Hardware Plan"
The UK government announced a forthcoming “AI Hardware Plan” aimed at capturing a 5% share of the global AI chip market. The initiative will debut at London Tech Week in June, backed by an initial £100 million ($125 million) fund for domestic...
Attack Path Visibility Concerns Outstrip AI Among Security Leaders
A recent study of cyber‑security leaders reveals that ransomware threats and the need for clear attack‑path visibility outrank the adoption of artificial‑intelligence solutions. Executives overseeing more than $100 billion in annual IT spend prioritize tools that map lateral movement and expose...
EY Hustles to Hire 'Forward Deployed Engineers'
EY, one of the Big Four professional services firms, announced it will initially recruit 45‑50 Forward Deployed Engineers (FDEs) to help clients move AI projects from prototype to production. The FDE role places technically skilled engineers directly within client organizations,...
US Government Has "Misspent" $3 Trillion. Auditors Want a Dedicated Unit to Tackle the Problem
The Government Accountability Office (GAO) estimates the U.S. government has misspent nearly $3 trillion since 2003, with improper payments hitting $185.8 billion in fiscal year 2025 – a $24 billion rise from the prior year. Overpayments, which represent 82% of the misspending, are...
Singapore and Latvia Punch Above Their Weight in NATO Cyber Battle
NATO’s annual Locked Shields cyber‑defense exercise this year centered on protecting national IT and OT infrastructure. Singapore, a non‑NATO member, again topped the competition, marking its second straight victory. The contest routinely invites external partners, allowing smaller nations like Singapore...
Streaming Architecture and Speculative Decoding: How Companies Are Unlocking Cheaper AI
Enterprises facing GPU scarcity and high power costs are adopting software techniques such as streaming architecture and speculative decoding to run AI models on cheaper hardware. These methods break inference into incremental token chunks and pre‑compute likely sequences, cutting compute...
OpenAI’s Hottest Models Are Coming to Bedrock
Amazon Web Services announced that it will integrate OpenAI’s flagship large language models into its Bedrock managed service within weeks, following a new partnership agreement. The deal gives AWS customers direct, API‑level access to models such as GPT‑4, GPT‑3.5‑Turbo and...
New Chinese Open Models Challenge Closed Western Top Tier
Chinese AI firms DeepSeek and Alibaba released DeepSeek V4 and Qwen 3.6‑27B last week, positioning them as open‑source alternatives to leading Western models. Both models deliver competitive performance while being freely accessible, undercutting the premium pricing of closed‑source offerings from OpenAI,...
STACKUP: The Stack's Weekly Tech Startups and Funding Wrap
The Stack’s weekly "STACKUP" newsletter curates the most recent startup financing rounds and milestone announcements across the tech ecosystem. This edition highlights over 120 deals that collectively raised roughly $3.2 billion, with a noticeable tilt toward AI‑driven platforms and fintech. Geographic...
“The CPU Is Reinserting Itself”, Will Reach Parity with GPUs for AI Workloads: Tan
Intel CEO Lip‑Bu Tan told analysts that CPUs are regaining prominence in AI, with the CPU‑to‑GPU inference ratio moving from 1:8 to 1:4 and heading toward parity. The chipmaker posted a $3.7 billion loss for Q1 2026 but beat revenue guidance,...
GitHub Bug Messed up Customer Code; COO Plays Down Incident
GitHub disclosed that an unreleased feature hidden behind an incomplete feature flag caused a merge‑queue bug that corrupted code in thousands of repositories on April 23. The defect, triggered when merge queues were combined with the squash‑merge method, led to...
China Now as Good as U.S. at Offensive Cyber: Dutch Intelligence
Dutch intelligence agencies warned that China’s offensive cyber capability now matches that of the United States. The AIVD’s annual report highlighted a sophisticated Chinese national program that evades detection, with only a tiny fraction of attacks being identified. Similar capabilities...
Disgruntled Claude Users Get an Explanation – and Another Option with ChatGPT-5.5
Anthropic addressed recent complaints from Claude users, attributing performance hiccups to a recent model update and resource throttling. At the same time, OpenAI unveiled ChatGPT‑5.5, which early testers say dramatically outperforms competitors on lengthy coding assignments. The new version reportedly...
The NCSC Is Worried About HDMI-Based Attacks, Should You Be?
The UK National Cyber Security Centre (NCSC) has unveiled SilentGlass, a hardware device that acts as an HDMI‑to‑DisplayPort firewall, blocking unexpected or malicious signals. Built on NCSC‑owned intellectual property and licensed to UK firm Goldilock, the product will be manufactured...
How Spotify Used Agents to Migrate 1,800 Data Pipelines and Save 10 Weeks of Dev Work
Spotify’s internal Honk tool deployed autonomous agents to migrate roughly 1,800 data pipelines across its backend. The system generated and applied code changes automatically, eliminating the need for manual rewrites. By the end of the effort, Spotify saved an estimated...
Apache ActiveMQ Bug Chain Gives Pre-Auth RCE, Is Getting Exploited
Researchers have confirmed active exploitation of a two‑step vulnerability chain in Apache ActiveMQ, combining CVE‑2026‑34197 (code injection) with CVE‑2024‑32114 (unauthenticated Jolokia exposure). The chain grants pre‑authentication remote code execution, and attacks have been observed in the wild, with canary hits...
Big Blue Says Its Sleeper AI Tool 'Bob' Boosted Dev Productivity 45%
IBM has quietly released its homegrown AI coding assistant, Bob, a secure on‑prem fork of VS Code that integrates large‑language‑model capabilities. Internal adoption lifted developer productivity by 45%, a figure highlighted by CEO Arvind Krishna during the Q1 2026 earnings call....