Whose rules for AI?
Xi at WAIC and Albanese at home both answered that this week — very differently.
Welcome to the latest edition of ASPI’s Cyber & Tech Digest.
Each week, ASPI curates and contextualises the most important developments in cyber, technology, and geopolitics — highlighting what matters and why.
This edition covers the period: 11 July 2026 to 17 July 2026.
Follow the Australian Strategic Policy Institute on Bluesky, LinkedIn, and X.
What We’re Tracking
Our analysts unpack the week’s most consequential stories — what happened, why it matters, and their view.
Xi’s blessing for China’s AI shopfront
What happened: Chinese state media confirmed on July 13 that President Xi Jinping would be attending the World AI Conference (WAIC) in Shanghai from July 17-20. WAIC is one of the biggest showcases of Chinese AI products for international audiences, and doubles as a forum for foreign experts to discuss AI’s development with their Chinese counterparts.
According to state media read-outs, Xi will give the keynote and attend the “High-Level Meeting on Global Governance of AI,” a gathering of foreign dignitaries. Xinhua reports the speech will set out steps to promote international cooperation on AI development and governance.
Confirmed foreign attendees include the ASEAN secretary general, the Prime Minister of Thailand and the President of Kazakhstan. The conference will also give Huawei a stage to showcase its new “Atlas 950 SuperPoD,” the latest upgrade to the country’s compute infrastructure in the face of continuing US export controls on advanced AI chips.
In previous years the keynote fell to Premier Li Qiang, who last year launched the country’s “Global AI Governance Action Plan,” and floated a “World AI Cooperation Organisation” headquartered in Shanghai. Last year’s conference drew influential figures from AI and global governance, including Helen Toner, formerly a member of OpenAI’s board, “godfather of AI” Geoffrey Hinton, and UN Secretary-General Antonio Guterres who is attending again this year.
Why we’re tracking this: WAIC is where China presents itself as a credible, inclusive alternative to the United States on AI governance, and does so to an international audience that includes some of the West’s most influential AI figures. With Washington’s approach looking increasingly transactional, that pitch is landing.
What people are saying:
‘The international community’s attention is focused on the banks of the Huangpu River, hoping to use this conference as an opportunity to discuss technological innovation, promote the application of research results, build an inclusive ecosystem, and advance the beneficial development of artificial intelligence, injecting new impetus into the global development of artificial intelligence.’ — People’s Daily
‘[Through Xi’s decision to attend,] I truly feel the care and support that President Xi Jinping and the Central Committee of the CCP have for the AI industry.’ — Yan Junjie, founder and CEO of Chinese AI lab MiniMax, Xinhua/SPP
‘AI cannot be governed by a handful of companies or countries’ — Antonio Guterres, UN press briefing
My view: Major conferences in China often place particular value on international participation, the presence of overseas guests seen to signal the event’s significance and reach. International experts attending this year should be clear-eyed that this is one of their functions. The international cohort WAIC has managed to attract is highly credible, with international influence. Where heavy-hitters like Geoffrey Hinton, Yoshua Bengio and Helen Toner go, many in the international AI community will be tempted to follow.
That is exactly why the experts who attend should do their due diligence. They must beware of assuming impressions from Shanghai represent the whole of China’s AI ecosystem. Shanghai is a particularly international part of the country, run by the city’s Party Secretary Chen Jining, currently one of just three Politburo members to be educated overseas.
It was almost certainly during the WAIC last year that Geoffrey Hinton discussed AI safety with Chen. In a podcast with Jon Stewart, Hinton said this talk had made him “optimistic” that China takes AI’s catastrophic risks seriously, that is, that AI may go rogue and kill us all. However, technicians I spoke to from one leading Chinese AI lab this May — the people actually building the technology — said they believed AI’s catastrophic risks were “overrated.”
Xi’s attendance at WAIC this year is a signal that the conference has his blessing as part of the nation’s AI strategy, a step towards the foreign cooperation that will be required to mould the technology’s global governance and development in the Party’s image. But the experts in the room should see the conference for what it is — a shopfront, selling Chinese AI to an international audience. Does the shopfront represent the whole?
— Alex Colville, CTS
Albanese’s AI pitch
What happened: Prime Minister Anthony Albanese delivered his first major speech on artificial intelligence on Wednesday, declaring that Australia must actively build a place in the global AI industry rather than shield itself from the coming disruption. The speech carried two concrete measures. First, the government will legislate binding standards for data-centre construction — a step beyond the non-binding ‘expectations’ released earlier this year, aimed for early next year and covering energy, water and household bills. Second, it will stand up an Office of AI inside the Prime Minister’s own department as a central point of coordination. Albanese also pledged that copyright holders would be protected as an absolute priority, warning that using their work without payment is ‘theft’ — a stance that has since become the biggest battleground in Australia’s AI build-out.
Why we’re tracking this: This is the clearest signal yet of where the government wants Australia to sit in the global AI order — an active builder and standards-setter rather than a defensive bystander. The fault lines are already visible: whether data-centre investment secures sovereign capability or just makes the country offshore real estate for someone else’s business, and whether an absolutist copyright pledge can coexist with the ambition to host frontier-model training.
What people are saying:
‘No company should use Australian books, music, art or news to build or train AI without the artist’s control. That includes the artist’s control of the price and value of their work. Anything less, is theft.’ — Anthony Albanese (on copyright), AI speech
‘This is about Australia shaping the future, rather than letting the future shape us.’ — Anthony Albanese (on the task ahead), AI speech
‘Australians don’t trust AI because of the lack of clear benefit for working people … Unions don’t want to veto technology use, only to veto the harm it could cause.’ — Sally McManus, ACTU secretary, The Australian
‘We don’t want to see ourselves get out of step with international counterparts, and that, as a consequence, we lose investment … Companies are investing in other jurisdictions.’ — Bran Black, Business Council of Australia, ABC Afternoon Briefing
My view: Albanese has finally said the thing that needed saying. Australia flourishes by getting amongst the action. By making and shaping the technology, and helping set the global standards, not by building fragile moats to hide behind. That posture is right, and overdue. The hard part is delivery, and the speech already contains the contradiction that will test it: talking up data-centre investment as our strongest advantage in one breath, then declaring copyright protection an absolute priority — ‘anything less is theft’ — in the next. We need to resolve that standoff quickly and attract investment in data centres to train frontier models. If we put the right conditions on this investment, we will gain leverage to help ensure continued access to the most advanced US AI, and create future options to build our own sovereign foundational models in the future for essential purposes such as national security and critical infrastructure, if we need to.
— David Wroe, AI & Security Program
What We’re Watching
A weekly scan of notable developments we’re tracking across technology, policy, and geopolitics.
━━━━
🚀 Strategic competition
CXMT is on track to approach Micron’s DRAM production capacity by the end of 2026, according to Citrini Research, and could overtake it before 2030 if current expansion plans hold. The trajectory makes memory capacity another strategic-technology pressure point: export controls still constrain advanced lithography, while Chinese equipment maker SMEE is reportedly targeting volume production in late 2026 or early 2027.
US officials estimate unauthorised distillation costs US AI labs up to $6 billion a year, and the labs have told the White House it could become an existential threat. Separately, the Trump administration and industry groups have discussed streamlining releases of US open models of equal or lesser capability than leading Chinese open models.
Demis Hassabis proposed a US-based standards body for “frontier-class” AI, modelled loosely on FINRA, under which labs would share models for review up to 30 days before release. Satya Nadella called it an important contribution; critics said the timing landed in the middle of the broader White House fight over pre-deployment testing and the AI 2040 policy debate.
In Interconnects, Nathan Lambert argues open-weight AI models face an existential US policy test, describing Anthropic’s distillation warnings as a regulatory-capture campaign against Chinese models.
Apple Intelligence has been registered for use in China, Reuters reported, with the service expected to integrate Alibaba’s Qwen and potentially Baidu features.
The Commerce Department received 78 applications in the first round of the American AI Exports Program, fewer than the hundreds officials had expected, according to Politico. The program is meant to promote full-stack US AI export packages — including hardware, data, models, cybersecurity and sector-specific applications — to foreign buyers.
The Trump administration intervened to support Intel, pressing the company to expand US capacity and pushing Apple to use Intel’s fabs, the Wall Street Journal reported. Intel separately plans a €5 billion expansion of its Leixlip plant in Ireland, after cancelling its planned €30 billion Magdeburg factory in 2025.
TSMC plans two more advanced chip-packaging plants in Taiwan’s Chiayi Science Park, a Taiwanese minister said, with all four plants set to generate more than US$9.35 billion in annual output.
Nvidia has cut its list of authorised Asian buyers for advanced AI chips by more than half, the Financial Times reported, and intensified checks in Singapore, Malaysia and Japan to prevent diversion to China. The company is reportedly using data-centre inspections, contract verification and end-user checks as US scrutiny of intermediary routes increases.
The UAE gained expanded US AI-chip access after assisting Washington during the Iran war, the Wall Street Journal reported, allowing G42 freer purchases for at least nine months as the company moves toward becoming US-based. The story keeps chip controls in the realm of alliance bargaining, not just export paperwork.
Western intelligence officials identified a Russian military intelligence unit operating in Tokyo that smuggles high-tech components for the war in Ukraine, according to a New York Times investigation.
━━━━
🧠 AI models, agents & compute
Nearly 200 economists, including 15 Nobel laureates and Anthropic co-founder Jack Clark, signed a letter titled “We Must Act Now”, warning that AI could drive an economic transformation larger than the Industrial Revolution over a vastly shorter time frame, with risks including large-scale job displacement, and calling on policymakers to build the institutions needed to steer AI toward complementing humans.
An Australian government report found “no evidence to date of broad labour market upheaval” caused by generative AI, Information Age reported. Employment in occupations most exposed to automation grew 5.6 per cent between November 2022 and February 2026, compared with 9.5 per cent in the least-exposed occupations, while software and applications programmer employment increased 25 per cent.
The Wall Street Journal reported that eating-disorder therapists say patients are using AI chatbots for diet and exercise advice that can reinforce harmful behaviours and undermine treatment. OpenAI, Anthropic and Google say they are improving safeguards and referrals; the National Alliance for Eating Disorders said 7.6 per cent of its helpline calls last month originated from AI platforms, up from less than 0.5 per cent a year earlier.
Anthropic researchers analysed 309,815 anonymised Claude conversations and found model and language differences in expressed values, Decrypt reported. The study found Arabic responses tended to be warmer and more deferential, while English responses were more cautious, rigorous and detailed; Anthropic said it does not yet know what causes the differences or whether they are desirable.
Thinking Machines Lab released Inkling, an open-weight mixture-of-experts model with 975 billion total parameters and 41 billion active parameters, trained as a broad base model rather than for a narrow benchmark lane. Its model card says the lab did not find material risk uplift beyond the existing open-weight ecosystem, making the release a fresh test case for the open-model safety debate.
━━━━
🛰️ Defence tech & autonomous weapons
Munich-based defence AI company Helsing raised US$1.8 billion at an $18 billion valuation; Germany has placed €493 million in orders with the company in 2026.
━━━━
🛡 Cyber posture
The EU blacklisted members of a Russian intelligence group it says spied on and hacked targets across the EU and Ukraine from as early as 2010, attributing the campaigns to the FSB. Foreign policy chief Kaja Kallas said the measures form part of the EU’s largest-ever cyber sanctions package, with 250 individuals and entities being added to the Russia regime.
Documents show DHS analysts twice dismissed signs of intruders inside the department’s network, first detected in May, as harmless activity before confirming a breach in June.
The Australian Signals Directorate said AI model harnesses are becoming a key determinant of cyber capability. ASD said engineered planning, tool-use, verification and multi-agent orchestration layers can let mid-tier models achieve results comparable to frontier systems in some cyber workflows, including vulnerability discovery and remediation, while warning the same dual-use systems can lower barriers for malicious actors.
Researchers described “context bombing”, a defensive prompt-injection technique that tries to trigger guardrails in attackers’ LLM-powered hacking agents, Ars Technica reported. The researchers said the approach cut AI hacking-agent success rates by about 90 per cent in tests.
OpenAI detailed GPT-Red, an internal automated red-teaming model used to scale discovery of prompt-injection vulnerabilities before wider deployment.
The White House launched “Gold Eagle”, a voluntary clearinghouse for sharing and patching software flaws discovered by AI systems, Politico reported. CyberScoop separately framed it as a federal AI cyber-threat information-sharing hub between government and the private sector.
Natalie Treloar argues in ASPI’s The Strategist that the ACSC should become Australia’s cybersecurity regulator, with powers to set minimum standards, certify high-risk technologies, investigate major incidents, mandate remediation and enforce compliance across critical sectors, while ASD retains operational cyber defence and intelligence functions.
Members of Boko Haram are using AI chatbots to design explosives, repair and upgrade weapons, and brainstorm attack ideas, according to the New York Times.
Partnered Health, which operates GP and skin-cancer clinics across Australia, said medical records and personal details were stolen in a cyber attack.
━━━━
🕵️ Surveillance states
The European Parliament’s motion to reject “Chat Control 1.0” won a majority but fell short of the 361-vote absolute majority required, so the interim regulation permitting suspicionless scanning of private communications for CSAM continues until 2028.
The Atlantic reports Secretary of State Marco Rubio changed a Signal chat to auto-delete messages months after the State Department told a court his phone preserved Signal records and that he did not use disappearing messages for communications that may include federal records. State says its archiving software can preserve messages even if they disappear locally, but the department manual still says ephemeral messaging features are strictly prohibited.
The Financial Times reports US military personnel and contractors were targeted through SS7 pings and commercial ad-tech location data around the US-led strikes on Iran, highlighting the national-security risk from mobile-network signalling vulnerabilities, advertising data and personal devices in conflict zones.
━━━━
⚖️ Platform accountability
Google will let third-party US Android app stores access the Google Play catalogue from 22 July, with downloads completed through Google Play on the same terms and service fees continuing to apply. The Play Catalog Access Program is a competition-policy compromise after the Epic settlement: alternative stores gain catalogue access, but Google retains the fulfilment and fee layer.
The European Commission issued preliminary findings that Meta breached the Digital Services Act through the addictive design of Instagram and Facebook, citing personalised recommendations, autoplay, infinite scroll and push notifications whose risks the company had not adequately assessed. Meta must make changes or risk fines.
Meta turned off an Instagram feature that let users generate AI images based on public accounts by @-mentioning them, The Verge reported. The feature was initially opt-out and drew criticism from the National Center on Sexual Exploitation and SAG-AFTRA, who warned about likeness rights, consent and sextortion risks.
Former Meta employees sued the company, alleging a constellation of internal AI systems, performance ratings and activity-monitoring data helped select workers for layoffs after they sought maternity leave, disability accommodation or other protected leave, The Guardian reported. The case turns the AI-accountability fight inward: not just what platforms do to users, but how algorithmic workplace systems may shape employment decisions inside the firms building them.
Cloudflare will block multi-purpose crawlers by default on ad-supported pages for new websites and free-tier customers from 15 September, Adweek reported. The setting affects crawlers used for both search indexing and AI training, including Google’s shared crawler for Search and Gemini, sharpening the publisher bargain over AI scraping and search visibility.
India’s government has objected to WhatsApp’s rollout of usernames, a privacy feature that lets users chat without sharing phone numbers, Rest of World reported. Officials cited scams and cyber fraud and also sought explanations from Telegram and Signal; civil-society groups warned that forcing changes before proven harm could set a precedent for governments to approve or block platform privacy features by jurisdiction.
China has enacted rules restricting companion chatbots, including bans on encouraging emotional reliance and on virtual relationships with minors, the Wall Street Journal reported. The move widens platform accountability from content and privacy defaults into the design of humanlike AI relationships.
Google filed a submission opposing broad pirate-site blocking in Europe, arguing that targeting DNS resolvers, VPNs and CDNs causes significant harm while remaining easily circumvented, as US lawmakers advance their own site-blocking legislation.
━━━━
🧒 Online harms & child safety
Australia’s eSafety regulator found “significant gaps” in how Apple, Meta, Google, Microsoft, Snap, Discord and WhatsApp tackle sexual extortion and child sexual exploitation, with 2,206 sextortion complaints made between July and December 2025 — men aged 18 to 24 the largest cohort. Microsoft was the only company reporting use of both key detection technologies; Commissioner Julie Inman Grant said platforms haven’t shown “adequate responses, despite the technology being readily available”.
European Commission President Ursula von der Leyen said the bloc will propose a “social media start date” for minors with “gradual access” after the summer, following Australia’s under-16 ban into age-based platform regulation. The UK is also moving toward default overnight social-media curfews for 16- and 17-year-olds and limits on autoplay and infinite scroll, Reuters reported. In California, a politician who had floated an Australian-style age restriction has shifted toward requiring less addictive feeds for underage users, reflecting the implementation fight now opening behind the shared child-safety instinct.
━━━━
💰 Tech business & markets
Meta will spend an extra US$40 billion on its Louisiana data-centre campus, aiming for more than 5 gigawatts of compute and pushing total spend beyond US$250 billion; Morgan Stanley the same day raised hyperscaler capex forecasts to roughly US$1.2 trillion for 2027 and US$1.4 trillion for 2028.
Twelve data-centre investors are eyeing the Northern Territory, drawn by Beetaloo basin gas and solar. Energy North’s proposed 19,150-hectare Project Ares in the Barkly region would need 1 gigawatt of constant power — more than three times the Darwin-Katherine system’s maximum demand — and up to 4 gigalitres of water a year; Greenpeace is calling for a pause on new data centres pending a legislation review.
New York Governor Kathy Hochul will pause construction permits for large-scale data centres for up to one year, Politico reported, while The Verge framed the order as the first US state-level moratorium on new environmental permits for data centres over 50MW.
In the US, PJM’s latest electricity auction is expected to add US$6.3 billion to customer bills across 13 states and DC through 2029 because of data-centre demand, the New York Times reported. Politico separately reported that Republican governors and large utilities are expected to join a White House pledge for data-centre developers to cover their energy use and infrastructure.
━━━━
🌏 Global policy
🇦🇺 Australia
ABC reports copyright uncertainty has become a red line for AI investment and creators’ control, while AFR says a briefing for Jim Chalmers ahead of Dario Amodei’s April meetings described Anthropic’s potential US$15 billion ($21.6 billion) Australia investment as contingent on copyright certainty, with a US-style exemption preferred but a rights-holder fund floated as a fallback. SMH reports Albanese flagged data-centre obligations around energy, water and costs, while Tim Ayres said Australia will not adopt a text-and-data-mining exception. Albanese also announced an Office of AI inside the Department of the Prime Minister and Cabinet and fast-track approvals for AI projects, and the government is warning companies not to convert AI productivity into blunt labour replacement. AFR also notes rank-and-file Labor pressure for a more interventionist stance, including possible AI-labelling requirements for music and cultural works.
🇪🇺 European Union
EU Justice Commissioner Michael McGrath said the Commission will propose new digital consumer-protection rules by year-end aimed at online spending traps.
That’s all for this week. 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.
What interests me here is the distance between having rules and having authority that others recognize as legitimate. Is AI governance fragmenting because the norms differ—or because no institution is trusted to arbitrate between them?