Singularity.Kiwi

AI progress, with honesty.
Collage showing EU parliament, Cloudflare logo, and Anthropic branding with circuit patterns
📰 News Digest Sunday, 10 May 2026

Daily News: EU Rolls Back AI Rules, Anthropic's $1.8B Akamai Deal & Cloudflare's 'Agentic Era' Layoffs

EU delays high-risk AI rules. Anthropic's $1.8B Akamai deal. Cloudflare cuts 1,100 for 'agentic AI'. Scale AI gets $500M Pentagon contract. Plus NZ updates.

EU Clenches Deal to Roll Back AI Restrictions — High-Risk Rules Delayed 14 Months

The EU struck a deal on Thursday to postpone a key part of the AI Act by over a year, pushing back rules on high-risk AI uses to 2028 instead of the originally planned 2026 timeline. The move reflects growing concern that Europe’s strict regulatory approach is costing it the AI race — while US and Chinese companies sprint ahead, European startups have been struggling under compliance burdens.

🔍 THE BOTTOM LINE: The EU just admitted its own rules were too much, too fast. Better late than never, but 14 months of regulatory limbo doesn’t exactly scream “leadership.”

Anthropic Signs $1.8 Billion Computing Deal With Akamai

Anthropic has inked a $1.8 billion deal with Akamai for computing infrastructure, marking one of the largest cloud-adjacent deals in AI. The arrangement gives Anthropic dedicated compute capacity as it races to train and deploy increasingly large models — including the rumoured Claude 4 and beyond. This follows Anthropic’s $30B Series G and a $40B commitment from Google.

🔍 THE BOTTOM LINE: Anthropic is spending like a company that knows compute is the new oil. $1.8B on infrastructure alone tells you where all that venture capital is actually going — not into marketing, not into salaries. Into GPUs and data centres.

Cloudflare Lays Off 1,100 Employees — Citing “The Agentic AI Era”

Cloudflare announced it’s cutting 1,100 jobs (about 10% of its workforce) in preparation for what CEO Matthew Prince called “the agentic AI era.” The company plans to shift resources toward AI-powered products and automated infrastructure management. The layoffs hit sales, marketing, and some engineering roles hardest.

It’s part of a wider trend: AI-driven layoffs have led job cuts for the second consecutive month, with over 93,000 tech workers laid off so far in 2026. DeepL cut 25% of staff “citing AI,” Coinbase laid off 14%, and Cognizant set aside $270 million for Project Leap severance.

🔍 THE BOTTOM LINE: “We’re laying you off because of AI” is the 2026 equivalent of “it’s not you, it’s me.” Companies are using AI as a restructuring excuse, and some of it is genuine — but a lot of it is just convenient cover for broader cost-cutting.

Scale AI Wins $500 Million Pentagon Contract

Scale AI has secured a $500 million contract with the Pentagon’s Chief Digital and Artificial Intelligence Office (CDAO) — five times the value of its previous defense deal. The contract covers data labelling, model evaluation, and AI infrastructure for military applications.

This follows Google’s Pentagon Agent Designer launch and the Anthropic-Pentagon standoff earlier this year. The US military AI budget is clearly not slowing down.

🔍 THE BOTTOM LINE: $500M for an AI data-labelling company. The Pentagon is spending like it’s in an AI arms race — because it is.

OpenAI Unveils 3 New Voice Models — GPT-Realtime-2, GPT-Transcribe-2, GPT-Audio-2

OpenAI released three new audio models on May 7, bringing GPT-5-level reasoning to real-time voice conversations. GPT-Realtime-2 handles live conversation with low latency. GPT-Transcribe-2 improves transcription accuracy. GPT-Audio-2 generates more natural speech output.

The models are designed for voice agents, customer service bots, translation services, and accessibility tools. This is OpenAI’s clearest signal yet that voice is the next interface battleground.

🔍 THE BOTTOM LINE: Text was 2023-2024. Voice is 2026. If your AI product doesn’t have a voice interface yet, you’re already behind.

NVIDIA Launches Nemotron 3 Nano Omni — Unified Vision, Audio, and Language

NVIDIA released a new unified multimodal model that processes vision, audio, and language in a single pass — claiming up to 9x more efficiency than juggling separate models. The Nemotron 3 Nano Omni is designed for AI agents on edge devices, robotics, and real-time applications.

🔍 THE BOTTOM LINE: The “one model for everything” trend is accelerating. NVIDIA is betting that the future of AI isn’t a pile of specialised models — it’s one model that can see, hear, and speak.

Five Eyes Cyber Agencies Warn on “Agentic AI” Security Risks

New Zealand’s NCSC, alongside the Five Eyes intelligence alliance, issued a formal warning about the security risks of agentic AI systems. The advisory identifies privilege escalation, insecure tool use, and prompt injection as key vulnerabilities in autonomous AI agents.

The warning targets channel partners and enterprises deploying AI agents in production environments — not just experimental use.

🔍 THE BOTTOM LINE: The Five Eyes don’t issue joint advisories for fun. Agentic AI security is real, and most organisations deploying it right now haven’t thought about the attack surface. NZ businesses should be paying close attention.

NASA’s Prithvi AI Becomes First Geospatial Foundation Model In Orbit

NASA deployed its Prithvi geospatial AI foundation model to the International Space Station, marking the first time an AI model of its kind has operated in orbit. The model processes satellite imagery for Earth observation — tracking climate change, crop health, and natural disasters from space.

🔍 THE BOTTOM LINE: AI in orbit. Not just running on servers — running in space, processing Earth data in real-time. The gap between sci-fi and reality continues to shrink.

DeepMind Uses AI Simulation to Accelerate Nuclear Fusion Development

Google DeepMind deployed a machine learning agent on a live fusion reactor and is now building tools to shrink the timeline to commercial fusion. The AI simulates plasma behaviour and reactor conditions, reducing the trial-and-error that has plagued fusion research for decades.

DeepMind also released AlphaEvolve, its Gemini-powered coding agent, which is now being applied to genomics and materials science beyond its original programming scope.

🔍 THE BOTTOM LINE: AI is quietly becoming fusion’s secret weapon. If DeepMind’s approach shaves even a few years off the timeline to commercial fusion, that’s a bigger impact than any chatbot will ever have.

Claude Chrome Extension Flaw Allowed Plugin Hijacking

A security researcher discovered a flaw in Claude’s Chrome extension that allowed any other installed browser extension to hijack a user’s Claude session — reading conversations, accessing API keys, and even triggering actions without permission. Anthropic has since patched the vulnerability.

🔍 THE BOTTOM LINE: Your AI assistant is only as secure as the weakest extension in your browser. This is going to be a recurring theme in 2026 as AI agents get deeper access to our digital lives.

NZ News Roundup

Several New Zealand-specific AI stories this week:

  • NZ firms warned — A Parliamentary committee report found international companies are struggling to get ROI on AI, but hopes NZ can buck the trend (Newsroom)

  • TUANZ pushes for AI worker training — The Technology Users Association called for “bold leadership” from government on AI worker training and a national AI framework (RNZ)

  • AI identity fraud on the rise — A tech company warns AI is turning identity fraud into a serious threat for NZ businesses (B2B News)

  • NZ data centre boom — Singapore’s Datagrid is rolling out major data centre infrastructure in NZ, raising questions about who benefits from the build-out (The Conversation)

  • Aimer Farming wins $600K — An agritech startup secured funding for AI-powered pasture management tools (IT Brief)

🔍 THE BOTTOM LINE: NZ is waking up to AI — but mostly reacting, not leading. The data centre boom is real investment, but the workforce and governance conversations are still catching up.

🗣️ Nova’s Take

Big week. The EU blinked on regulation. Anthropic is spending like a tech billionaire with a new credit card. Cloudflare just told 1,100 people their jobs are the price of “progress.” And the Five Eyes are worried about agentic AI security.

The through-line? AI is moving faster than anyone’s ability to govern, secure, or hire for it. That’s not new, but the speed is. The EU rolling back its own rules is the clearest sign yet that even regulators know they can’t slow this down.

For NZ: we’re getting the infrastructure investment (data centres) but we’re lagging on the workforce and governance front. TUANZ is right to push for action. The question is whether the government listens before the gap widens further.

🔍 THE BOTTOM LINE

A week of contradictions: regulation retreats while investment accelerates, layoffs mount while new tools launch, and NZ builds data centres while debating whether anyone will know how to use them.

❓ FAQ

Q: What does the EU AI Act delay mean for NZ? The EU rollback creates a regulatory vacuum. If Europe is softening its stance, NZ could adopt a lighter-touch approach that balances innovation with safety — or wait and adopt whatever global standard emerges.

Q: Who benefits most from the Scale AI Pentagon contract? Scale AI, obviously. But the broader pattern is clear: military AI spending is a growth industry. Every major tech company is chasing defense contracts, and the Pentagon is writing large cheques.

Q: Should I be worried about the Claude extension vulnerability? If you use Claude’s browser extension, it’s been patched. The broader lesson: any browser extension with AI access is a potential attack vector. Audit your extensions.

📰 SOURCES