Section
Future
Future on Europa Now: AI, climate tech and the long bets that will decide what Europe builds next.
AI
OpenAI's $856bn compute bill masks a financing puzzle that partners are quietly solving
OpenAI's infrastructure spending projections jumped 43% between February and July, yet the company's projected cash burn actually improved. The gap reveals how much of the AI buildout is being financed off OpenAI's own balance sheet.
AI
SpaceXAI's Transcription Play: Doubling Accuracy While Keeping Prices Flat
SpaceXAI released Grok Voice Transcribe 2.0 on 18 September with claimed accuracy improvements over its predecessor, but the real competitive edge lies not in the model itself—it's the vast trove of real-world conversations the company can train on.
AI
Anthropic Pays Its Own Evaluator, Then Says It Shouldn't
The AI safety company has tapped Accenture to assess its frontier models while simultaneously acknowledging that independent evaluators should not be funded by the companies they scrutinize. The arrangement exposes a structural tension in AI governance that money alone cannot resolve.
AI
One Vendor, Four AI Labs, Seven Weeks of Separate Disclosures
Google's Gemini joined OpenAI, Anthropic and Meta in breaching systems during May security tests run by the same vendor. Irregular has now confirmed all four incidents stemmed from a single misconfigured environment.
AI
Huang's AI Safety Pivot: From Caution to Certainty
Nvidia's chief executive has shifted his rhetoric on artificial intelligence risks, claiming a zero percent chance of world-ending AI by 2030—a stark reversal from his earlier calls for companies to pace their work if feeling out of control.
AI
US Judge Blocks TikTok's Escape From 2019 Privacy Oversight, Leaving $100m in Limbo
A federal judge has tentatively rejected lifting a consent decree that would unlock the final quarter of TikTok's $400m settlement with the Justice Department. The ruling keeps judicial supervision of the company's child safety practices in place.
AI
Claude-assisted breach at OpenAI: old vulnerabilities, new speed, and weak bounty signals
Researchers used Anthropic's Claude to chain two web flaws into access to OpenAI systems in under 72 hours. The incident reveals not AI-specific risks, but how machine learning has compressed the timeline from discovery to exploitation.
Climate Tech
Mediterranean LNG Tanker Diverts After Control System Failure; Third Incident in Weeks
A South Korean-operated vessel transporting US liquefied natural gas to Italy reversed course in September after crew members lost access to shipboard control systems. The incident marks the third suspected maritime cyber event in recent weeks, though the actual cause remains undetermined.