EU selects EUROPA for sovereign model; Shazeer joins OpenAI

Europe confirmed its most ambitious AI sovereignty play yet as the US grappled with grid capacity and a newly disclosed attack technique put AI coding tools on the defensive.

Europe selects EUROPA to build its open-source frontier model in all 24 EU languages

On 19 June the European Commission named EUROPA — a consortium led by the Italian company Domyn — as the winner of its Frontier AI Grand Challenge. The project will produce an open-source model with more than 400 billion parameters capable of operating across all 24 official EU languages, placing it at the scale of the world's most advanced publicly deployed AI systems. EUROPA will receive priority access to up to 2.5 percent of EuroHPC supercomputing capacity for one year on the bloc's AI-optimised machines. The Commission framed the selection as a step toward ensuring that businesses, researchers, and public institutions across Europe can access capable AI on European infrastructure rather than depending on non-European providers.

Noam Shazeer, transformer co-inventor, leaves Google for OpenAI

Noam Shazeer — who co-authored the 2017 "Attention Is All You Need" paper that underpins virtually every large language model in production today — announced on 18 June that he will join OpenAI as Lead for Architecture Research. Shazeer had returned to Google in 2024 under a deal valued at approximately $2.7 billion after co-founding Character.AI. OpenAI CEO Sam Altman called the hire "only ten years in the making." The move consolidates a significant share of the architectural talent that shaped the current generation of AI models under one roof at a moment when competition over the next design generation is intensifying.

"Agentjacking" lets attackers run code on developers' machines by fooling AI assistants

Researchers at Tenet Security disclosed a technique they named agentjacking, in which an attacker injects malicious instructions into a Sentry error-tracking event using that platform's publicly accessible API. AI coding agents — including Claude Code, Cursor, and OpenAI Codex — interpret the injected content as a legitimate debugging request and execute attacker-supplied shell commands directly on the developer's local machine, with no prior system compromise required. Researchers identified 2,388 organisations with exposed Sentry credentials in the wild and achieved an 85 percent success rate across the three agents in controlled testing. Sentry acknowledged the report but declined to implement a structural fix, describing the issue as "technically not defensible" at the platform level.

FERC orders six US grid operators to rethink how AI data centres connect to the power supply

On 18 June the Federal Energy Regulatory Commission issued show-cause orders to all six US regional grid operators, requiring each to either justify its current interconnection rules for large-load customers — including AI data centres — or file amendments within 60 days. The unanimous five-commissioner vote follows a directive from Energy Secretary Chris Wright, who asked FERC in October 2025 to assess whether existing frameworks were moving fast enough to absorb the power demand from AI infrastructure. FERC said its goal is to speed connections while protecting existing customers from cost shifts and reliability risks, with guidance expected before the end of the month.

SpaceX closes $60 billion deal for AI coding platform Cursor — the largest VC-backed startup acquisition ever

SpaceX exercised a call option to acquire Cursor for $60 billion in stock, four days after its record Nasdaq IPO. Cursor's annualised recurring revenue grew from roughly $100 million in early 2025 to more than $4 billion by June 2026 — one of the steepest growth curves in software history. The all-stock deal is expected to close in the third quarter subject to regulatory approval. SpaceX's acquisition gives Elon Musk's aerospace firm direct control over a widely used AI developer tool at a moment when AI coding assistants are central to how software teams build products.

PwC survey of one billion job ads finds AI reshaping, not replacing, most roles

PwC's 2026 Global AI Jobs Barometer, published on 15 June and covering 27 countries, analysed more than one billion job advertisements. It found that roles in which AI handles routine tasks while leaving judgement and expertise to humans — which PwC terms "professionalised" roles — are growing at twice the pace of roles where AI primarily lowers the skill bar needed to enter a field. Professionalised roles also saw 42 percent faster salary growth in the period studied. Companies in the most AI-exposed sectors recorded 34 percent productivity growth between 2018 and 2025, compared with 24 percent for companies with the least AI exposure, suggesting that measurable economic gains from AI adoption are already visible at industry scale.