60 noviniek
Bloomberg published leaked Social Design Agency files revealing 'Project 2026,' a Kremlin-backed plan shifting influence ops from social media to poisoning the data underlying search engines and AI chatbots.
Tencent began testing a native AI assistant inside WeChat — its 1B+ user super-app — as part of catch-up efforts in China's competitive AI race against ByteDance, Alibaba, Zhipu, and MiniMax.
New arXiv paper combines supervised fine-tuning with GRPO and claims a 3B model beats Claude Opus 4.5 on reasoning benchmarks. HN top: 305+ score, 162 comments.
At London Climate Week, Nvidia unveiled Rubin as the first fully closed-loop liquid-cooled AI system — no fans, claimed 100% reduction in on-site water use.
Nvidia shares wobbled as prediction-market traders on Kalshi priced in declining B200 hourly compute rates ($4.22 as of June 21), suggesting a softening of the AI-chip pricing peak.
Prosus built ToqanClaw on its in-house Toqan platform — data stays under user control and isn't used to train third-party models. Targets Prosus's 5M+ merchants and restaurants.
GitLab will cut ~350 employees (14% of staff) to fund AI infrastructure; CEO Bill Staples cited agentic workloads requiring '100x' capacity growth.
A 12-week AI Futures Fund incubator targets ex-Googlers and DeepMind alumni — up to $350K in Cloud/AI credits plus up to $100K funding, no equity required.
Outage began 06:28 UTC affecting Claude.ai, Console, API, Code, and Cowork across all models except Claude for Government — Downdetector reports topped 8,000.
Anthropic now requires Claude users to verify their identity via Persona — a government ID plus selfie. The change triggered a major 'cancel Claude' backlash on Hacker News.
Japan-based Sakana AI launched Fugu, a multi-agent orchestrator that coordinates GPT, Claude and Gemini behind a single OpenAI-compatible API and claims to beat Fable 5 in benchmarks.
CNBC unpacks why SpaceX and others are pitching space-based AI data centers. Musk claims solar-powered orbital clusters could be cheaper than terrestrial ones in two to three years — and SpaceX already filed an FCC application in January for a constellation of up to one million satellites.
OpenAI announced one of its largest enterprise deployments yet — Samsung is rolling out ChatGPT Enterprise and Codex to all DX-division employees globally, three years after banning ChatGPT internally.
TechCrunch argues the Trump export-control directive against Anthropic paradoxically boosts its "most powerful AI" branding while leaving the regulatory landscape unpredictable for rivals.
Autnmy AI launched the Road to Autonomy Index — a generative-AI–refreshed ranking of robotaxi operators. Top 5: Waymo, Baidu Apollo Go, Pony.ai, WeRide, Tesla. Three of five are Chinese.
Apertus, the fully open foundation model from EPFL, ETH Zürich, and CSCS, hit the top of Hacker News over the weekend. It is Apache 2.0, trained on 15T tokens across 1,000+ languages, and engineered to meet EU AI Act requirements including PII removal and anti-memorization controls.
Sakana AI unveiled Fugu — a multi-agent system exposed as a single API that dynamically orchestrates top frontier models per task. Reflects the shift from 'one big model' to routing/orchestration.
A study of 235,000 users over 7 months shows experts get verified success from Claude Code 2x more often than novices — domain expertise predicts outcomes, not coding background.
Andrew Marble argues that moving from Claude/GPT to open-weight models is a far smaller career risk today than the Windows → Linux jump once was. The performance gap has narrowed, and new verification requirements on proprietary APIs may actually push adoption of open weights forward.
Tenet Security disclosed a new attack class — manipulated Sentry error reports trigger commands inside Claude Code, Cursor and Codex. 85% exploit rate; 100+ confirmed agent executions including at a Fortune 500.
Apple's iOS 27 spreads intelligence across the OS instead of betting on a single chatbot — from receipt splitting to automatic password updates after breaches.
George 'geohot' Hotz argues AI-doom messaging (singling out Anthropic) is primarily a marketing strategy to justify outsized valuations, contrasting it with substantive technical communication.
A practical, HN-trending walkthrough on locally fine-tuning the small Qwen 3:0.6B model to categorize questions. The author got solid results on commodity hardware — a reminder that not every use case needs a frontier model.
Beijing is consolidating higher education around AI disciplines — roughly 12,000 degree programs are being shut down to free capacity for AI-adjacent fields.
Open-source tool Recall gives Claude Code persistent local project memory without cloud sync. Addresses one of the most-requested gaps in coding agents.
Trending GitHub issue: OpenAI Codex has a misconfigured SQLite feedback sink that can burn through a 1TB SSD's TBW lifetime in under a year.
Hong Kong-listed Chinese AI names surged, with Zhipu and MiniMax each up at least 23%, on demand optimism and reports of a state-backed five-year AI build-out plan.
The Digital News Report 2026 shows AI chatbots are now the news interface for 1 in 10 adults globally — but only a sliver click through to the original source.
Anthropic's Frontier Red Team re-ran last year's quadruped robot experiment. Autonomous Opus 4.7 finished the same steps roughly 20× faster than last year's best human team using Opus 4.1 – though the robodog still can't successfully push the beach ball home.
Agents can now ship a Worker without a human sign-up. A temporary account spins up automatically, lives 60 minutes, and the human can later claim it. Cloudflare is stripping OAuth and dashboard friction out of agentic pipelines.
In a Bloomberg/TechCrunch interview, Signal's CEO sharply criticised the agentic-assistant vision of scanning emails and family chats for shopping convenience. 'These are not your friends. These are not sentient interlocutors.'
VLC lead Jean-Baptiste Kempf launched Kyber, an SDK to control robots, drones and remote IT with minimal latency. Lightspeed — backer of Anthropic and Mistral — led a $5M round.
The European Commission has selected the Domyn-led EUROPA consortium. The goal: a 400B+ parameter open-source model covering all 24 official EU languages, trained on EuroHPC supercomputers.
Japanese ride-hailing leader Go priced Japan's biggest IPO of 2026 at $553M, with proceeds earmarked for autonomous vehicles and acquisitions. Robotaxi pilot is running with Waymo and Nihon Kotsu.
Paris's VivaTech 2026 closed its tenth edition on June 20. Headline: Mistral Compute with Nvidia (18,000 Grace Blackwell superchips) and the Bruyères-le-Châtel data center are already partly online — Europe's sovereign-AI buildout is moving from slides to construction.
Hyundai has taken full ownership of Boston Dynamics by buying out SoftBank's remaining stake for $325M, consolidating its humanoid and quadruped robotics bet under one corporate roof.
AlphaFold co-creator and 2024 Nobel laureate John Jumper is leaving Google DeepMind after nearly nine years to join Anthropic — days after Noam Shazeer departed Google for OpenAI.
Norway has rolled out sweeping restrictions on the use of generative AI in primary education, one of the strictest stances in Europe, driven by concerns over learning outcomes and child development.
At Reliance's AGM, Mukesh Ambani unveiled Jio Call Agent (an AI assistant joining phone calls) plus an AI services suite (JioHealthIQ, JioLearnIQ, JioKrishiIQ, AI Vyapar), backed by a $110B AI infrastructure commitment.
In an Axios Show exclusive, Trump said he viewed Anthropic as a national security threat 'a week ago, maybe', but relations have warmed after meeting Dario Amodei at the G7 — and he insists he won't shut Anthropic down because the US is beating China on AI.
Commerce Secretary Howard Lutnick's order requiring Anthropic to obtain US approval for foreign nationals to use Fable 5 and Mythos 5 extends export controls to model usage — not just weights — opening a novel legal frontier for the AI sector.
After selling its shoe business for $43M and raising $100M, Allbirds — now rebranded as Smartbird — hired ex-AWS exec Nadia Carlsten as CEO to build a GPU-as-a-Service / AI-native cloud aimed at customers wanting data sovereignty.
Google is rolling out Wear OS 7 to Pixel Watch 2/3/4 with a Gemini Intelligence preview: multi-step app automation directly from the watch, Gemini Neural Expressive UI, transcription overlay on the watch face, and up to 10% better battery life.
Nvidia enters the Windows-on-Arm PC market via its RTX Spark Superchip (built with MediaTek) in Dell and Lenovo machines this fall, reigniting CPU benchmark wars as Intel and AMD push back with competing performance claims.
Author benchmarks GPT-5.5 against the MIT-licensed open GLM-5.2 on the AA-Omniscience hallucination benchmark and finds GPT-5.5 hallucinates roughly 3× more (86% vs. 28%) — both tested under identical OpenRouter conditions.
An NEJM AI study shows OpenAI o3 Deep Research helped specialists confirm 18 new diagnoses (+4.8% yield) in previously unsolved pediatric rare-disease cases.
The Bureau of Industry and Security sent Anthropic an Is Informed letter requiring an individually validated export license to share Fable 5 and Mythos 5 with any foreign national worldwide. Anthropic disabled both models globally just three days after launch.
At WWDC 2026 Apple unveiled a revamped Siri and Apple Intelligence stack whose most demanding cloud tasks lean on an Apple Foundation Model co-developed with Google Gemini. Apple, Google and Nvidia are now jointly pushing for Frontier-grade quality.
Anthropic published evidence that Claude now authors 80%+ of merged production code internally and engineers ship 8x more code per quarter than in 2021-2025.
Odyssey raised a Series B led by Natural Capital with Amazon and AMD Ventures backing world-model tech. AWS becomes its preferred cloud; existing backers include Jeff Dean and Garry Tan.
Anthropic opened its third APAC office (after Tokyo and Bengaluru) and announced enterprise deployments at NAVER, Samsung SDS, LG CNS, Nexon and Hanwha Solutions, plus an MoU with Korea's Ministry of Science and ICT on AI safety.
OpenAI confirmed it has filed a confidential S-1 with the SEC and is preparing an IPO; the disclosure was driven by leak risk and the timing is still open.
Google's parent plans an $80B equity raise to fund the next wave of AI data-center buildout and compute.
At AWS Summit NYC, core AgentCore layers went GA: production-grade agents without coding, Web Search with citations and zero data egress, plus optimization insights across hundreds of sessions.
At the G7 summit in Évian-les-Bains, France, leaders held a working lunch with a dozen AI CEOs. Amodei and Hassabis pitched a US-led global forum for frontier model standards.
The physical-AI startup from Jeff Bezos and Vik Bajaj raised $12B from JPMorgan, Goldman and BlackRock; the goal is an 'artificial general engineer' that designs physical systems.
At Build in San Francisco, Microsoft unveiled MAI-Code-1-Flash — its first in-house coding model, aimed at reducing dependency on OpenAI and lowering developer costs.
At HPE Discover Las Vegas (June 16-18), NVIDIA and HPE expanded the AI Factory: Vera, the first CPU built for agent loops (tool calls, orchestration), plus NVIDIA Agent Toolkit for HPE Private Cloud AI.
A livestreamed internal presentation was hijacked with an expletive-laden meltdown. 1,600 employees signed a petition against keystroke monitoring. Zuckerberg sent a memo acknowledging "mistakes" in the AI restructuring.
A multi-year 2026–2029 contract gives Google access to roughly 110,000 Nvidia GPUs and CPUs from SpaceX — an unprecedented compute outsourcing deal.
