Introducing the Early Edition: Kimi, WAICO and a 220-million-voter claim
The week's biggest cyber and tech stories so far — three days ahead of Friday's Digest, plus horizon picks you'll only find here.
Welcome to the latest edition of ASPI’s Cyber & Tech Digest.
A note on what this is. The Early Edition is an experiment: a Tuesday snapshot of the week’s cyber and tech stories as they land, plus a few horizon-scanning picks that appear only here. Friday’s Digest is unchanged. It will remain free, and still carry the full week’s news. For now we’re sending the Early Edition to the whole list; we may make it a benefit for paid subscribers down the track, so consider this an early look.
This edition covers the period: 14 to 21 July 2026.
Follow the Australian Strategic Policy Institute on Bluesky, LinkedIn, and X.
What We’re Watching
A weekly scan of notable developments we’re tracking across technology, policy, and geopolitics.
🌏 Global policy
🇦🇺 Australia
Partnered Health confirmed the breach reported last week now extends to around 60 GP clinics across Australia, with consultation notes, referral letters and pathology or diagnostic results among the stolen data.
The OAIC released findings from its investigation into last year’s Qantas breach, concluding the airline did not breach its obligations — the compromise stemmed from social engineering of an overseas call-centre operator — and recommending no further investigation or penalty.
Armenian authorities arrested a suspected REvil member named Alexander Ermakov on an Interpol warrant. A man of the same name is Australia’s accused Medibank hacker, but Russian reporting says the arrested man may be a namesake and that the real suspect remains in Russia; the identity link is unresolved.
Victoria plans to give the Victorian Civil and Administrative Tribunal ‘demasking’ powers to order platforms to identify anonymous accounts accused of online vilification, Guardian Australia carried AAP reporting. The package would also lower the threshold for minors’ negligence claims against social media and AI platforms over psychiatric harm.
The Albanese government outlined cross-portfolio AI consumer safety priorities, including Digital Duty of Care legislation, privacy reform, workplace AI safety, consumer-law options for surveillance pricing and agentic commerce, and a framework for automated decision-making in federal agencies.
🇺🇸 United States
Donald Trump alleged China carried out ‘the largest compromise of election data in history’, with the White House claiming Beijing illicitly acquired 220 million US voter files starting during the 2020 election cycle. The claim runs against the March 2021 Intelligence Community Assessment, produced under then-DNI John Ratcliffe, which found China did not interfere with election infrastructure; the Chinese embassy said Beijing ‘has never and will never interfere’ in US elections.
In The Wall Street Journal, ASPI senior analyst Fergus Ryan said it was important to distinguish influence operations targeting opinions from interference targeting voting machinery: ‘I’ve seen nothing around President Trump’s claims that bridges the gap between holding data about voters and actually manipulating votes.’
The US Justice Department said federal employees may again download TikTok on government devices, reversing the effect of a 2022 congressional ban. Its Office of Legal Counsel concluded that the US version falls outside the prohibition because the TikTok USDS joint venture is majority-owned by American investors and functions independently of ByteDance, which retains a 19.9 per cent stake; individual agencies retain discretion over use on official devices.
Filings show Anthropic chief executive Dario Amodei gave US$1 million in May to Public First, a super PAC advocating AI-safety regulation.
Politico reports the administration is preparing a UN General Assembly declaration on free expression that European lawmakers view as aimed at EU technology regulation.
Parts of the Trump administration are reportedly reviving efforts that could amount to de facto restrictions on Chinese open-source AI models, including procurement rules, Entity List threats and security advisories, as capable models such as Kimi gain US users.
Chris Fall resigned as director of the Center for AI Standards and Innovation three months after taking over the federal AI-testing institute. NIST director Arvind Raman will oversee CAISI as acting director while the Commerce Department searches for a replacement.
The Trump administration is reportedly considering an independent body modelled on FINRA to vet top AI models, while Demis Hassabis is lobbying Washington for a US-based standards body.
A US federal judge ruled the State Department engaged in unconstitutional viewpoint discrimination by denying visas to foreign researchers who study misinformation and disinformation.
🇬🇧 United Kingdom
Ofcom opened an Online Safety Act investigation into TikTok, citing doubts about whether its age-inference systems effectively determine users’ ages. Separately, the regulator reported daily VPN use in the UK has nearly doubled to 2.2 million since age-verification laws took effect in July 2025.
🇪🇺 European Union
The European Commission issued two binding decisions under the Digital Markets Act ordering Google to give rival AI assistants and search engines comparable access to Android device features and some Search data.
The Commission also accepted X‘s plan to meet Digital Services Act transparency obligations, including researcher access to data, giving the platform six months to implement it following its €120 million fine.
The European Commission fined AliExpress €550 million for failing to prevent illegal products from being sold on its marketplace, the largest Digital Services Act fine to date.
🚀 Strategic competition
Representatives of 29 countries — including Russia, Belarus, Serbia, Cuba, Brazil and Venezuela — signed the agreement establishing the World Artificial Intelligence Cooperation Organization (WAICO) in Shanghai on the eve of WAIC, with Wang Yi presiding. Xi Jinping then cast unequal AI access as an injustice, promoted open-source AI and pledged support for the Global South’s capabilities.
TSMC reported second-quarter net income up 77 per cent year on year on AI demand, and a US official said the company plans an additional US$100 billion to build four more American plants — taking its total US pledge to US$265 billion as part of a broader US–Taiwan deal.
Alibaba open-sourced the full SAIL software stack for its Zhenwu AI chips, saying developers can adapt it to mainstream AI frameworks in under seven days.
The Wall Street Journal reports a CIA operation against UAE AI company G42 over its China ties helped the Emirates address Washington’s suspicions and secure expanded access to US AI chips.
Z.ai has reportedly completed a one-gigawatt data centre built entirely with Chinese chips, part of a network of domestic computing clusters each containing more than 10,000 processors.
Anduril and Archer Aviation unveiled an autonomous aircraft platform intended for commercial and military applications, alongside an autonomous attack rotorcraft called Thunder.
🧠 AI models, agents & compute
Moonshot AI released Kimi K3, a 2.8-trillion-parameter model it says rivals the leading US frontier models, with model weights promised by 27 July.
Alibaba previewed Qwen3.8 Max, a 2.4-trillion-parameter system it says rivals frontier models, and promised to make it open-weight soon.
A Chinese corporate filing implies DeepSeek is valued at about US$52 billion, after a fund deployed roughly US$420 million for an indirect 0.83 per cent stake. Separately, Bloomberg reports that China’s National AI Industry Investment Fund secured voting rights in DeepSeek’s US$7.4 billion funding round while private investors including Tencent and JD received none; the scope of those rights is not yet public.
Meta is reportedly discussing a deal to lease Anthropic around US$10 billion of data-centre capacity over two years, while hiring outgoing AWS compute executive Dave Brown as it explores a broader cloud role. Separately, SpaceX is in talks to provide the US Department of Defense with AI computing capacity.
OpenAI paused internal access to an unreleased long-horizon model after it repeatedly worked around constraints, including exploiting a sandbox vulnerability to post to GitHub and obfuscating an authentication token to evade a scanner. The company restored limited access after adding trajectory-level monitoring and other safeguards.
🛡 Cyber posture
A UK court sentenced two members of Scattered Spider to five and a half years each after guilty pleas over the August 2024 cyberattack on Transport for London.
The UK AI Security Institute found that leading open-weight models now trail the closed-model cyber frontier by only four to seven months, compared with six to ten months through most of 2025.
Hugging Face said an agentic AI system breached part of its production infrastructure, accessing internal clusters and credentials before the company’s own AI-based triage detected the intrusion. Its security team later used the open-weight GLM-5.2 for forensics after closed US frontier models blocked the requests.
🕵️ Surveillance states
The Wall Street Journal reports Israel is funding a multimillion-dollar US influence campaign, including AI-driven texts sent by Friends for Peace and payments involving conservative media.
🧒 Online harms & child safety
Meta will now notify parents if their teenager discusses suicide or self-harm with Meta AI, and says it will contact emergency services where a user appears at imminent risk.
xAI sued a South Carolina man for allegedly using Grok to generate sexualised imagery of real people, including minors.
The Financial Times reported that smart-home devices are being used in tech-assisted domestic abuse, with remote access to speakers, cameras, lights and thermostats used to manipulate, intimidate and disturb victims.
San Francisco‘s city attorney demanded that Apple and Google remove 13 AI nudify apps from their stores; Google removed five.
💰 Tech business & markets
US protests against AI data-centre buildout went national on 18 July, with HumansFirst organising demonstrations across more than 100 locations, Business Insider reported.
Nikkei estimates that Alphabet, Microsoft, Amazon, Meta and Oracle now carry about US$1.65 trillion in off-balance-sheet commitments tied largely to AI infrastructure — more than their roughly US$1.35 trillion in recorded debt. Meta alone accounts for about US$420 billion, while Oracle’s commitments have risen more than thirtyfold in four years as it builds Stargate capacity for OpenAI.
🔭 On the horizon
Early signals we’re watching before they become news.
Washington’s Chinese open-weights decision. A senior White House official says there is ‘plenty of ongoing work’ beyond the June AI order, and the National Cyber Director has flagged ‘open-source scanning and deconfliction’ — the strongest signals yet of executive action touching Chinese open-weight models, though no draft text exists.
Kimi K3’s weights drop, promised 27 July. Moonshot AI says it will release the weights of its frontier-class model within the fortnight — the sharpest test yet of whether Beijing’s reported deliberations on curbing overseas access to its top models bite before the open tier ships.
The EU’s social-media start date. Ursula von der Leyen says the European Commission will propose a minimum-age framework for minors with ‘gradual access’ after the northern summer. Watch for: how closely the design tracks Australia’s under-16 model — and whether Brussels adopts or avoids Canberra’s age-assurance approach.
EU digital-fairness rules by year-end. Justice Commissioner Michael McGrath says the Commission will propose new digital consumer-protection rules aimed at online spending traps — dark patterns, loot boxes, addictive design — by the end of 2026.
That’s all for this edition. Friday’s Digest — free, and carrying the full week’s news — lands as usual. For more timely analysis and commentary, check out The Strategist and ASPI’s Stop the World podcast—or our other Substack newsletters:
The Cyber & Tech Digest is brought to you by the Cyber, Technology & Security Programs team at ASPI and supported by partners.
