Washington's clampdown on frontier AI is starting to look like a strategic own-goal: Chinese lab Z.ai's open-weight GLM-5.2 reportedly now matches Anthropic's Mythos on some cybersecurity tasks, even as the Trump administration edges toward letting Anthropic's powerful Fable 5 back online — and Austria openly lobbies for the EU to host Anthropic on European soil. Europe is leaning into the opening on its own terms, with Germany casting AI as a €300-billion fix for its worker shortage, a draft EU Innovation Act that would let public bodies bankroll product innovation, and deeptech investment already hitting an annual record halfway through the year. Beneath the optimism, the Bank for International Settlements warned that an AI-investment bust now ranks among the gravest threats to global prosperity as capital pours through lightly regulated non-bank channels. On the threat side it was a quieter day, though a flawed third-party email platform exposed up to 14.2 million accounts across six Japanese ISPs at KDDI, ShinyHunters folded food giant Sysco into its sprawling Salesforce-linked extortion wave, and fresh analysis detailed how North Korea's UNC1069 used weeks-long AI-generated personas to poison the "axios" npm package and brush against OpenAI's code-signing.
Top Stories
- China Has Matched Anthropic in Cybersecurity, Resetting AI Race — Technology - WSJ.com · AI & Power
- Austria Lobbies EU to Host Anthropic After US Access Curbs — Bloomberg Technology · EU & Technology
- Scoop: Powerful Anthropic model, Fable 5, on track to return soon — Axios · AI & Power
- German AI Rollout Offers €300 Billion Fix for Worker Shortage — Bloomberg Technology · EU & Technology
- GPT-5.6: The System Card — Don't Worry About the Vase · AI & Power
AI & Power
China Has Matched Anthropic in Cybersecurity, Resetting AI Race — Technology - WSJ.com
Why it matters: A Chinese open model reportedly matching a US frontier system on offensive-cyber tasks turns the AI-export clampdown into a strategic own-goal.
Researchers say Z.ai's open-weight GLM-5.2 matches Anthropic's Mythos in some bug-finding and cyber scenarios, fuelling concern that curbing top US models hands Beijing a cyberwarfare edge.
Scoop: Powerful Anthropic model, Fable 5, on track to return soon — Axios
Why it matters: The on-again, off-again gatekeeping of Anthropic's most capable model shows Washington now decides, case by case, who gets frontier AI.
The Trump administration is close to letting Anthropic restore access to its powerful Fable 5 model, offline for 15 days over government security fears.
GPT-5.6: The System Card — Don't Worry About the Vase
Why it matters: The GPT-5.6 system card is the clearest signal yet of where frontier capability — and its cyber-risk guardrails — now sit.
An analysis of OpenAI's GPT-5.6 system card parses the new flagship's capabilities and safety stack ahead of a still-restricted general release.
AI Bust Risks Ripple Effects From Growth to Credit, BIS Says — Bloomberg Technology
Why it matters: When the central banks' bank flags an AI bust as a top systemic threat, the bubble debate moves from markets chatter to financial-stability risk.
The Bank for International Settlements warned that an AI investment bust — alongside inflation and fiscal stress — is among the most alarming threats to global prosperity, with capital flowing through loosely regulated non-bank channels.
EU & Technology
Austria Lobbies EU to Host Anthropic After US Access Curbs — Bloomberg Technology
Why it matters: A scramble to host a US frontier lab on EU soil is digital sovereignty made literal — Europe sensing opportunity in Washington's export curbs.
Austria is lobbying for the EU to host Anthropic after US access curbs on its Claude Mythos model, pitching Europe as an alternative home for frontier AI.
German AI Rollout Offers €300 Billion Fix for Worker Shortage — Bloomberg Technology
Why it matters: Germany framing AI as a EUR300bn answer to its labour crunch is the clearest European bet yet on adoption over regulation.
German firms' AI adoption could add some EUR300 billion to the economy and offset a deepening worker shortage, from automating invoice processing to broader operations.
New EU law will allow public bodies to finance product innovation — Tech Archives | Euractiv
Why it matters: An EU Innovation Act letting public bodies bankroll product innovation is Brussels reaching for industrial-policy levers in the competitiveness race.
The draft EU Innovation Act would let public bodies finance product innovation and strengthen startups' intellectual-property position, part of the bloc's competitiveness push.
European deeptech investment hits annual record in just six months — Sifted
Why it matters: Record deeptech funding in half a year is a rare bright spot for European tech sovereignty.
European deeptech investment has already hit an annual record just six months into 2026, signalling momentum in the continent's hard-tech and sovereignty-relevant sectors.
Exclusive: British Business Bank to back UK scaleups with £400m per year — Sifted
Why it matters: State capital for scaleups is the UK trying to fix the growth-stage funding gap that sends its best startups abroad.
The British Business Bank will commit GBP400 million a year to back UK scaleups, targeting the growth-stage funding gap.
Europe’s strategic autonomy starts on the road — Policy – POLITICO
Why it matters: Strategic autonomy, resilience and economic security are now the organising frame for EU tech and industrial policy.
Geopolitical tension, energy volatility and economic-security worries have pushed competitiveness, resilience and strategic autonomy to the centre of EU policy.
China & Technology
Peking University and DeepSeek Open-Source DSpark, Delivering Major Leap in LLM Inference Efficiency — Pandaily - China Tech News, AI & Electric Vehicle Insights
Why it matters: An open-sourced inference breakthrough from China's top lab-and-university pairing chips away at the West's efficiency edge.
Peking University and DeepSeek open-sourced DSpark, a speculative-decoding framework that boosts LLM inference efficiency by 60-85%, easing chip strain.
Reading Beijing’s Signal to Brussels — Pekingnology
Why it matters: A state-media commentary read as Beijing's message to Brussels signals where China wants the EU trade-and-tech relationship to go.
A CCTV-linked commentary is parsed as Beijing's deliberate signal to Brussels ahead of EU-China trade talks, framing the desired terms of the relationship.
China Achieves 100% Domestically Produced Nuclear Fusion Superconducting Magnets in 'Artificial Sun' Breakthrough — Pandaily - China Tech News, AI & Electric Vehicle Insights
Why it matters: Fully domestic fusion-magnet production is another brick in China's drive for end-to-end strategic-tech self-sufficiency.
China says two independently developed superconducting-magnet systems for fusion reactors passed technical acceptance, reaching 100% domestic production in an 'artificial sun' milestone.
China's Aerospace-Grade Carbon Fiber Breaks Free from Foreign Dependency with Major Production Expansion — Pandaily - China Tech News, AI & Electric Vehicle Insights
Why it matters: Domestic aerospace-grade carbon fibre removes another Western chokehold over China's high-end manufacturing.
Zhongfu Shenying inaugurated three high-performance production lines, advancing China toward self-sufficiency in aerospace-grade carbon fibre.
China's Photonic Computing Answer to Space: Lighter, Cooler, Radiation-Immune Alternative to Traditional Chips — Pandaily - China Tech News, AI & Electric Vehicle Insights
Why it matters: Photonic, radiation-immune computing is China hunting an alternative to GPUs constrained by export controls.
Guangbenwei Technology is developing photon-based computing for space - lighter, cooler and radiation-immune - as an alternative to traditional chips.
Space Solar Power Station Written into China's 15th Five-Year Plan, Three Technical Paths Mapped Out — Pandaily - China Tech News, AI & Electric Vehicle Insights
Why it matters: Writing space-based solar into the Five-Year Plan moves a moonshot from research into state-backed engineering.
China's 15th Five-Year Plan formally includes a space solar power station for the first time, mapping three technical paths from research toward engineering.
Defence & National Security
Agentic-AI tool aims to give US commanders new target options ‘within seconds’ — Defense One - All Content
Why it matters: Putting agentic AI into the targeting loop is the governance question of military AI made urgent.
A US agentic-AI tool aims to give commanders new target options 'within seconds', even as concerns persist over the power and governance of military software agents.
UK Plans ‘Hybrid’ Warship to Replace Aging Destroyer Fleet — Bloomberg Politics
Why it matters: Swapping prestige destroyers for cheaper 'common combat vessels' is the attritable-warfare turn reaching the surface fleet.
Britain scrapped plans for a new destroyer, opting for at least six cheaper 'common combat vessels' to ready the navy for modern, attritable warfare.
Threat Intelligence (CTI)
[P2] Manufacturing Trust: How UNC1069 Used AI Personas to Breach the Open-Source Supply Chain — Threat Intelligence on Medium
Why it matters: North Korea weaponising AI-generated personas to subvert an open-source maintainer - and reach OpenAI's code-signing - is the supply-chain threat's next form.
Analysis of UNC1069 (North Korea-nexus) using multi-week AI-generated personas to socially engineer an npm maintainer and publish backdoored 'axios' package versions (WAVESHAPER.V2 RAT); the poisoned package even executed in a GitHub Actions workflow tied to OpenAI's macOS code-signing.
severity high · exploited in the wild · EU: NIS2, AI Act · actor UNC1069 (North Korea nexus) (75%)
Cybersecurity & Threats
[P2] Data breach exposes up to 14.2 million email logins at six ISPs — BleepingComputer
Why it matters: A 14-million-account breach at a major telecom, via flawed third-party email software, is a textbook supply-chain exposure of essential infrastructure.
A vulnerability in third-party software in KDDI's managed email platform exposed up to 14.22 million email accounts (hashed/encrypted passwords) across six Japanese ISPs (Nifty, Biglobe, JCOM, STNet, Chubu Telecom, KDDI Web), detected 17 June.
severity high · exploited in the wild · EU: NIS2
[P1] Sysco - 2,691,852 breached accounts — Have I Been Pwned latest breaches
Why it matters: Another Sysco extortion - this time ShinyHunters - shows the crew's Salesforce-linked 'pay or leak' wave still rolling through global enterprises.
ShinyHunters' 'pay or leak' extortion published 2.7M unique email addresses (staff and customers - names, phones, addresses, job titles) from US food distributor Sysco; the group also claims 61M Salesforce records, and Qilin ransomware hit Sysco weeks earlier.
severity high · exploited in the wild · EU: GDPR · actor ShinyHunters (70%), escalation