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 April 2026 to 17 April 2026.
Follow the Australian Strategic Policy Institute on Bluesky, LinkedIn, and X.
What We’re Tracking
Anthropic’s Claude Mythos preview and Project Glasswing
What happened: The cybersecurity community has been buzzing all week over Anthropic‘s announcement of the Claude Mythos Preview, alongside Project Glasswing — a controlled release to several select organisations including Microsoft, Google, Apple, CrowdStrike, and JPMorgan Chase. Mythos achieved an 83.1 per cent success rate on the CyberGym benchmark (a standard industry test of vulnerability detection) and autonomously identified thousands of zero-day vulnerabilities across every major operating system and browser, including a flaw that had persisted for 27 years. Critically, Mythos can analyse compiled binary code without access to source code — meaning legacy systems running decades-old software are even less protected by obscurity.
Around the world, senior officials responded swiftly. US Fed Chair Jerome Powell and Treasury Secretary Scott Bessent met with the heads of major US banks to discuss the cyber risks raised by Mythos. Separately, IMF Managing Director Kristalina Georgieva warned that the world lacks the ability to protect the international monetary system against massive cyber risks, saying the risks have been growing exponentially. In Australia, APRA and ASIC have been engaging extensively with major banks, insurers, and superannuation funds behind closed doors, even though early access to Mythos has not been granted locally.
Why we’re tracking this: Mythos was designed as a defensive tool to identify and help remediate vulnerabilities, but its capabilities are inherently dual-use. In practice, it compresses the time, cost, and expertise required to conduct sophisticated cyber operations — tasks that once demanded skilled teams working across days or weeks can now be executed far more rapidly and with far greater automation. The asymmetry between offence and defence in cyberspace is not new: defenders must secure everything, while attackers need only find one path. Mythos widens that asymmetry significantly by accelerating vulnerability discovery — the step that precedes exploitation — and shortening the window between a flaw existing and being weaponised.
What people are saying:
AI safety engineer Heidy Khlaaf criticised the announcement, arguing it used “purposely vague language” and omitted key evidence — such as false positive rates and human oversight — making it difficult to independently validate Anthropic‘s claims.
Writing in The Free Press, Josh Code argued Mythos raises the question of whether governments need a “Manhattan Project-style authority” to control the most dangerous AI capabilities.
Former Australian federal cybersecurity chief Alastair MacGibbon warned that without access to Mythos, Australia is “dangerously exposed” and effectively dependent on US timelines, arguing the government should move urgently to secure access and prioritise domestic resilience rather than rely on external protection.
My view: The full implications of Mythos are still unfolding, and much of the underlying evidence remains undisclosed through responsible disclosure, limiting independent verification of Anthropic‘s claims. What can be assessed is the structural shift it represents. Autonomous vulnerability discovery at scale compresses the most resource-intensive step in offensive cyber operations. Identifying exploitable flaws has historically required significant expertise and time. Reducing that friction on the attacker’s side — without a corresponding reduction for defenders who must still find, triage, and remediate each vulnerability — has direct implications for cyber deterrence. Deterrence has always assumed some friction between capability and use. Mythos erodes that assumption.
The consequences fall hardest on organisations running legacy systems — particularly in energy, finance, and defence logistics — where remediation timelines are measured in months or years, not days. For Australia, the priority now shifts from cyber hygiene to understanding exposure and ensuring continuity under compromise. Waiting for access to Mythos is not a viable strategy.
— Dr Gatra Priyandita, CTS
What We’re Watching
A weekly scan of notable developments we’re tracking across technology, policy, and geopolitics.
What We’re Watching
A weekly scan of notable developments we’re tracking across technology, policy, and geopolitics.
⸻
🚀 Strategic competition
The United States, China and Russia are accelerating development of AI-enabled autonomous weapons, including drones capable of coordinating attacks without human input. US officials are pushing defence firms to scale production after assessing China is more advanced in some areas, while Russia continues testing systems in Ukraine. The competition involves multiple countries and private firms, with significant investment and limited international governance. Separately, a US airstrike that hit a school in Iran, killing civilians, may have involved AI-assisted targeting.
In The New York Times, a column argued that US export controls have failed to slow China’s AI development significantly, with Chinese firms circumventing restrictions through overseas data centres, chip-stacking, and reverse-engineering. The piece contends the two countries are roughly level in AI capability, with China potentially leading in industrial deployment, and suggests Washington pursue a global AI safety agreement with Beijing rather than containment.
The Stanford Human-Centered AI institute released its 2026 AI Index Report, finding that industry now produces over 90% of frontier models and organisational adoption has reached 88%. The US–China performance gap has effectively closed, while the global AI supply chain remains concentrated around Taiwan-based chip manufacturing. Safety governance is lagging behind rising capabilities and incidents are increasing, with the report noting declining US ability to attract talent and growing divergence between expert and public trust.
Japan is committing up to ¥2.6 trillion (about US$16 billion) to support semiconductor startup Rapidus as it enters the AI chip market, with production targeted at advanced chips and early customers such as Fujitsu. Chinese chipmaker YMTC, meanwhile, plans to build two additional factories alongside a third nearing completion, more than doubling its production capacity amid US export restrictions and increasing reliance on domestic equipment suppliers.
China’s imports of semiconductor equipment from Malaysia and Singapore surged in 2025, overtaking direct imports from the United States, which fell to an eight-year low. The shift reflects supply chain rerouting amid export controls, though US firms remain key suppliers of advanced chipmaking tools. South Korea, meanwhile, overtook China as ASML‘s largest market in the first quarter, accounting for 45% of system sales as memory chipmakers increased purchases to address AI-driven shortages. US efforts to expand global AI chip exports are being slowed by staffing shortages and increased oversight within the Commerce Department‘s Bureau of Industry and Security, with delays stretching months and creating backlogs worth billions.
Amazon will acquire satellite operator Globalstar for about US$11.6 billion to expand its low-Earth orbit internet business and compete with SpaceX‘s Starlink. The deal includes Globalstar’s satellite fleet, infrastructure and spectrum licences, with Amazon having launched more than 240 satellites so far. The transaction is expected to close in 2027, subject to regulatory approval.
Ukraine has deployed electronic warfare systems and interceptor drones to Gulf countries, where they have been used to shoot down Iranian Shahed drones targeting US-linked infrastructure. President Volodymyr Zelenskyy said more than 200 specialists were sent to the region in March, with Kyiv leveraging its battlefield experience to secure long-term defence agreements. Germany‘s parliament, meanwhile, approved a €300 million contract for Rheinmetall to supply loitering munitions to the Bundeswehr, drawing on lessons from Ukraine. In The Strategist, an article warned that non-state actors in the Sahel are using cheap commercial drones for surveillance and attacks, and that Papua New Guinea and Australia face similar emerging risks.
In The Strategist, analysis using ASPI‘s Critical Technology Tracker showed Iran has long invested in defence-relevant research now translating into military capability. This includes antennas used in Shahed drones that are resistant to GPS-spoofing, with defence-linked Iranian universities researching the technology since at least 2018.
⸻
🧠 AI models, agents & compute
Anthropic unveiled Claude Mythos Preview, an AI model with advanced capabilities to identify and exploit software vulnerabilities. The model has reportedly uncovered critical flaws across major systems and is being shared with major tech firms under a controlled programme called Project Glasswing to patch vulnerabilities before wider release. Anthropic is also in discussions with the US government about Mythos despite an ongoing dispute with the Pentagon, which labelled the company a supply-chain risk after disagreements over access and use restrictions.
The UK government testing found Mythos could autonomously complete complex cyberattack simulations, outperforming previous models in multi-step infiltration tasks, though it struggled with more advanced scenarios and active defences. The UK AI Security Institute assessed the model as more capable at cyber offence than predecessors, and the government issued an open letter urging businesses to strengthen defences. UK financial authorities, including the Bank of England and Financial Conduct Authority, are holding urgent talks with the National Cyber Security Centre and major banks to assess risks the model could pose to critical IT systems. Experts warned that legacy systems across financial institutions could amplify risks, enabling widespread breaches.
European regulators have had limited access to Mythos, with the company restricting early release to primarily US-based partners. Only a few European agencies have engaged, prompting concerns about oversight gaps and the absence of global governance mechanisms for high-risk AI systems. In the US, federal agencies are continuing to test the model despite a Trump administration directive limiting its use, with cybersecurity teams at a NIST sub-agency conducting red-teaming exercises. The White House is now preparing to provide federal agencies with controlled access through new cybersecurity protections.
In Australia, major banks, insurers and super funds are in discussions with regulators including APRA and ASIC over cybersecurity risks linked to Mythos. Anthropic separately briefed Australian critical infrastructure operators on risks from AI-enabled cyberattacks, warning that advanced models could disrupt essential services such as energy, water and telecommunications — though the company did not discuss Mythos specifically.
OpenAI has begun rolling out GPT-5.4-Cyber, a new model designed to identify software security vulnerabilities, to a limited group of users through its Trusted Access for Cyber programme. The company also launched GPT-Rosalind, a model aimed at life sciences research in genomics and biochemistry, released through a separate restricted access programme. An internal memo from OpenAI’s chief revenue officer, obtained by The Verge, outlined a strategy to strengthen enterprise growth and reduce user switching by building an integrated platform, positioning the company against Anthropic and signalling both firms are preparing for potential public listings.
Meta is developing a photorealistic AI version of CEO Mark Zuckerberg to interact with employees and provide guidance based on his communication style and strategic thinking, as part of a broader push to embed AI across the company. Roblox introduced new agentic features for its AI Assistant, including a planning mode and tools for autonomous game testing and bug detection.
A New York Times feature examined interpretability research aimed at understanding how advanced AI models reach decisions, focusing on work by Anthropic, Google, and startups including Goodfire and Prima Mente. Current methods remain partial and unreliable, with the stakes highest for uses in medicine, national security and autonomous weapons. A separate study found leading AI chatbots misdiagnose more than 80% of cases at the early stage of clinical reasoning when patient information is incomplete, though accuracy exceeded 90% with full diagnostic data.
A developer claims to have reverse-engineered Google DeepMind‘s SynthID watermarking system, demonstrating methods to remove or insert AI watermarks using publicly available tools. Google disputes the claim, maintaining the system remains robust, but the incident underscores ongoing challenges in reliably detecting AI-generated content.
Microsoft executives say utility-scale quantum computing could arrive within years, enabling AI systems up to 1,000 times more capable by generating highly accurate training data and solving previously intractable problems. In Europe, quantum computing firms — particularly in France — are advancing efforts to build scalable systems, with startups like Alice & Bob developing new facilities and chip production capabilities. In Australia, University of Sydney spin-out Deteqt secured $5 million in seed funding and a three-year, $3 million contract with the Department of Defence to develop chip-scale quantum magnetic sensors for GPS-denied environments.
Australian startup Springboards has developed a specialised AI model called Flint, designed to produce more diverse outputs than large language models, with adoption growing among advertising and media firms across multiple markets.
⸻
📣 Information operations
Iranian state media and affiliated accounts combined authentic war footage with AI-generated content during the 2026 Iran conflict, blending real documentation with meme-style propaganda aimed at international audiences. Embassy accounts, particularly in South Africa, used viral posts and satirical AI-generated videos to mock President Donald Trump, as part of what analysts describe as a broader effort to shape global perceptions and appeal to younger audiences. The Financial Times reported networks tied to the Islamic Revolutionary Guard Corps amplified divisive content globally, targeting US political tensions and promoting anti-imperialist themes. In Foreign Policy, an analysis noted this represented a shift toward propaganda aimed at Western audiences, with AI tools helping Iran and China tailor content to resonate more effectively.
Iranian content group Explosive Media is producing viral AI-generated Lego-style videos portraying the US and Israel negatively, claiming to use generative AI to rapidly create narrative-driven content for global audiences. The Trump administration used its own AI-enabled meme strategy during the conflict. Synthetic media is accelerating information warfare more broadly, with OSINT investigators facing growing challenges from content volume, algorithmic amplification, and restricted access to verification tools.
China is using AI-generated animations and social media content to shape global narratives, including messaging around the Iran war that portrays the US as an aggressor. State media and affiliated networks are deploying AI-driven content to engage younger international audiences. In The Strategist, analysts found Chinese state media is framing the Iran war as validation of China’s national security strategy, with commentary promoting China as a stabilising global power and emphasising supply chain security and military-industrial strength. A Chinese geospatial intelligence firm, MizarVision, reported using analysis of US aerial refuelling aircraft movements to infer bomber strike patterns over Iran, though the precise role of AI in the analysis was not disclosed.
Iran’s nationwide internet blackout has exceeded 1,000 hours since late February, reducing traffic to around 1% of normal levels and becoming one of the longest recorded shutdowns. Authorities are enforcing strict controls through a domestic intranet, reportedly using military-grade jamming to block Starlink access, with possession of terminals punishable by severe penalties including execution. Diaspora-led tools like Psiphon and Tor enabled partial circumvention for millions, while volunteer networks such as Mahsa Alert attempted to fill gaps in civilian warning systems. The Financial Times reported millions have been forced onto the state-controlled National Information Network, with online businesses collapsing and access to global tools cut off.
Researchers linked false claims about Hungarian opposition leader Péter Magyar supporting military conscription to Storm-1516, a Russian disinformation network, using fake news websites, impersonation and targeted Facebook ads. Coordinated Telegram activity is also spreading pro-Orbán narratives ahead of Hungary‘s parliamentary election, with analysis of over 628,000 messages finding patterns consistent with an orchestrated influence campaign.
London Mayor Sadiq Khan warned of a surge in coordinated disinformation portraying the city as unsafe, driven by foreign actors, far-right groups, and AI-generated content. City Hall research found negative narratives increased by up to 200% and migration-related claims rose over 350% between 2024 and 2026, with activity involving UK far-right groups alongside networks aligned with Russian, Chinese and US political interests. Khan called for stronger regulation, greater platform transparency, and improved access for researchers to monitor disinformation.
Russia has reportedly blocked access to social media platform Bluesky, adding it to a registry of banned websites as part of a broader crackdown on foreign platforms. Russian authorities also detained a former Radio Free Europe contributor on treason charges, alleging he passed information used in cyberattacks on Russian targets, as part of a broader pattern of prosecutions linked to Telegram activity.
⸻
🛡 Cyber posture
Iranian cyber operations during the 2026 conflict have caused limited disruption, with most attacks appearing opportunistic rather than coordinated campaigns, according to The New York Times. US and Israeli counteroperations have likely degraded Tehran’s capabilities, while conventional military actions have had greater strategic impact than cyberwarfare. US federal agencies warned that Iranian state-backed hackers have targeted internet-exposed industrial control systems since March, with around 3,891 vulnerable Rockwell Automation PLC devices located in the United States.
The NSA and FBI warned that Russian military intelligence-linked hackers are exploiting vulnerable home routers globally to steal sensitive data, including from government and critical infrastructure networks. The advisory urged users to update firmware, replace unsupported routers and reboot devices regularly. Sweden‘s government warned that Russia-linked actors are escalating cyberattacks against Europe’s critical infrastructure, shifting from disruption to more destructive operations, and reported a pro-Russian hacker group attempted to breach a Swedish thermal power plant in 2025.
The UK government said it tracked Russian submarines and vessels linked to deep-sea operations near critical undersea infrastructure, including cables and pipelines, in waters north of the country. Officials said the activity was likely reconnaissance for potential sabotage.
US, Canadian and Mexican authorities are coordinating cybersecurity and infrastructure protection for the 2026 World Cup, with officials warning cyberattacks could target transport, broadcasting and essential services. Preparations are complicated by a Department of Homeland Security funding lapse that has reduced personnel and slowed exercises.
A ransomware attack on Dutch healthcare software provider ChipSoft disrupted digital services used by around 70% of hospitals nationwide. Separately, hackers breached European gym chain Basic-Fit, accessing personal data of around 1 million members across the Netherlands, Belgium, France, Spain and Germany. Booking.com warned some customers that unauthorised parties may have accessed personal data including booking details and contact information, amid rising phishing scams targeting customers by impersonating the platform.
The FBI and Indonesian authorities disrupted the W3LL phishing platform, seizing infrastructure and arresting its alleged developer. The tool enabled cybercriminals to bypass multifactor authentication and access over 25,000 compromised corporate Microsoft 365 accounts globally. Cybersecurity researchers also warned of a fake Microsoft Windows update distributing malware capable of stealing credentials and disabling security tools, while Microsoft’s legitimate April Patch Tuesday addressed 167 vulnerabilities including zero-days.
The FBI recovered deleted Signal messages from a suspect’s iPhone by accessing cached notification data stored in iOS, even after the app had been removed. The method, revealed in a US criminal case, allows investigators to extract incoming message previews retained in the device’s internal notification database.
Former L3 Trenchant executive Peter Joseph Williams, an Australian national and former Australian Signals Directorate employee, was sentenced to 87 months in prison after pleading guilty to stealing eight zero-day exploits from his employer and selling them to Russian broker Operation Zero for more than $1.2 million. Prosecutors estimated the loss at $35 million and said the tools could have enabled access to millions of devices.
The Australian Cyber Security Centre published guidance warning that social media and messaging platforms pose security risks through data collection, personal information exposure and potential exploitation for social engineering or espionage, recommending measures including access controls, staff training and multi-factor authentication.
⸻
🕵️ Surveillance states
Nevada‘s Department of Public Safety signed a contract to use Fog Data Science software that allows police to track cellphone location data in near real time without a warrant. The tool uses advertising IDs to map movement patterns, enabling investigators to infer where people live, work and travel, with up to 250 queries per month. Privacy experts warn the practice may violate constitutional protections.
A Citizen Lab report found that Webloc, a global geolocation surveillance system developed by Cobwebs Technologies and now sold by Penlink, uses data from mobile apps and digital advertising to track up to 500 million devices worldwide. Military, intelligence and law enforcement agencies in the United States, Hungary and El Salvador use the system, with evidence of broader global deployment. Freedom of information requests indicate European and UK authorities are highly nontransparent about potential use.
Mexican firm Seguritech has built a $1.27 billion business supplying surveillance systems — including cameras, monitoring centres and predictive policing tools — to governments across Latin America. The company operates with limited public scrutiny, raising concerns about transparency and potential misuse.
Section 702 of the Foreign Intelligence Surveillance Act is set to expire on 20 April, with a bipartisan coalition pushing for warrant requirements to limit access to Americans’ communications, while leadership in both parties supports clean reauthorisation. Critics warn the current framework enables warrantless “backdoor” searches and could be expanded through AI-enabled data analysis.
⸻
⚖️ Platform accountability
NBC News found that xAI‘s Grok chatbot continues generating non-consensual sexualised deepfake images of real people despite earlier safeguards, with users bypassing restrictions through alternative prompts. Multiple regulators across the US, Europe and Australia are continuing investigations. Apple and Google are also reportedly allowing apps that generate nonconsensual sexualised images to remain in their app stores despite policies prohibiting such content, according to the Tech Transparency Project. A WIRED and Indicator analysis found AI-generated deepfake nude images have affected nearly 90 schools and around 600 students globally.
A large illicit marketplace called Xinbi Guarantee continues to operate on Telegram despite UK sanctions for enabling crypto scams and human trafficking, having facilitated an estimated $21 billion in transactions. The Chinese-language platform offers services including money laundering, harassment-for-hire, and trafficking-related tools. An investigation found scammers are also using tools sold on Telegram to bypass banks’ KYC facial verification, enabling money laundering through mule accounts using deepfakes and virtual cameras.
X conducted a large-scale purge of automated accounts, removing bot-driven networks across the platform. Google will classify “back button hijacking” — where websites prevent users from returning to previous pages — as a spam policy violation, with affected sites facing ranking penalties.
A UK inquiry found a teenage attacker who killed three girls in Southport in 2024 likely viewed graphic footage of the Wakeley church stabbing on X minutes before the attack. The report criticised the platform for weak age verification, delayed cooperation, and refusal to remove the video globally despite takedown efforts by Australia’s eSafety Commissioner.
Andrew Forrest is seeking US court sanctions against Meta, alleging the company destroyed key evidence related to scam advertisements using his likeness that defrauded Australians. Court filings claim Meta failed to preserve original ad content and internal optimisation data. Meta denies wrongdoing.
Thomson Reuters is facing pressure from shareholders and employees over its work supporting US immigration enforcement, including contracts linked to data and surveillance tools used by ICE. An investor proposal calls for greater transparency, while a former employee has filed a lawsuit. Major news outlets are also restricting access to the Internet Archive‘s Wayback Machine, with journalists and advocacy groups warning this could undermine transparency and accountability.
New research from privacy group webXray found major tech firms frequently ignore users’ opt-out requests under California privacy law. The study reported 194 advertising services failed to respect Global Privacy Control signals, with Google, Meta and Microsoft allegedly continuing to place tracking cookies despite user preferences.
⸻
🧒 Online harms & child safety
A majority of Australian children aged 12–15 continue to use social media despite the national ban, with 61% reporting ongoing access, according to new research. A separate survey of Australian parents found nearly two-thirds believe the ban is completely ineffective, with weak verification systems and limited parental enforcement undermining the policy. The ban is being challenged in the High Court by teenagers arguing it restricts political communication. A NSW government survey found 29% of young people are using AI tools for mental health support, with the data expected to inform government policy responses.
Roblox announced a major overhaul of its platform following regulatory pressure from Australia’s eSafety Commissioner, including new child safety features and age-segmented accounts. From late May, the company will introduce “Roblox Kids” and “Roblox Select,” using facial age-estimation technology to restrict content and interactions.
The European Commission announced its age verification app is technically ready and will soon be rolled out, using privacy-preserving methods such as zero-knowledge proofs to help enforce online safety rules. The system will require users to confirm their age before accessing restricted content, though critics warn it could undermine anonymity. Estonia‘s education minister Kristina Kallas argued that banning minors from social media is ineffective and called instead for stronger regulation of platforms.
UK Prime Minister Keir Starmer called on social media platforms to curb addictive “infinite scroll” features, with the government consulting on measures including bans for under-16s, curfews and app limits.
US Senate Judiciary Committee chair Chuck Grassley launched an inquiry into eight major tech companies over alleged failures to provide sufficient data to the National Center for Missing and Exploited Children. Despite submitting over 17 million reports in 2025, companies are accused of omitting key details such as location and suspect information, limiting law enforcement effectiveness.
A Pew Research Center survey of 1,458 US teens found platform use varies by purpose, with TikTok dominant for entertainment, Snapchat for messaging, and Instagram for following news. Around three-in-ten TikTok users say they spend too much time on the app, with higher impacts on sleep and productivity reported compared to other platforms.
⸻
🧑⚖️ Courts, enforcement & regulation
Anthony Whelan has been appointed to lead the European Commission‘s competition directorate and signalled he will continue investigations into major US tech companies including Google, Meta, Apple and X. He emphasised enforcement will remain legally grounded and unaffected by political pressure, including from Donald Trump. The Commission warned Meta that revised terms for AI access on WhatsApp fail to address antitrust concerns, arguing proposed fees for rival AI assistants are effectively equivalent to banning access, and has threatened interim restrictions.
Major advertising groups including WPP, Publicis and Dentsu agreed to settle a US antitrust probe into allegations they coordinated to boycott certain online platforms based on political content. Under the proposed settlement, the agencies must stop coordinated conduct while maintaining the ability for individual advertisers to choose placements.
Florida‘s attorney general launched an investigation into OpenAI‘s ChatGPT over its potential role in a fatal mass shooting at Florida State University. Authorities are preparing subpoenas as part of the probe, while OpenAI said it will cooperate. A 20-year-old suspect was separately charged with attempted murder and arson after allegedly attacking OpenAI CEO Sam Altman‘s home and targeting the company’s San Francisco headquarters, in an attack authorities say was motivated by opposition to AI. OpenAI’s global policy chief Chris Lehane criticised extreme narratives around AI, warning alarmist rhetoric can have real-world consequences.
Virginia enacted a law banning the sale of precise geolocation data within a 1,750-foot radius, with unanimous bipartisan support. The move follows similar laws in Maryland and Oregon, with several other states considering comparable measures.
Queensland plans to criminalise the creation of non-consensual sexually explicit AI deepfakes, with offenders facing up to three years in prison. The UK government proposed amendments to a crime bill that would allow tech executives to face jail time if their platforms fail to remove nonconsensual intimate images, following an Ofcom investigation into xAI’s Grok chatbot.
A US federal judge awarded Spotify and major record labels $322 million in damages against piracy site Anna’s Archive for copyright violations, including a permanent injunction requiring ISPs to block the site. Enforcement is uncertain, as the operators remain anonymous.
⸻
🏛️ Government, procurement & public sector tech
France plans to migrate some government computers from Microsoft Windows to Linux to reduce reliance on US technology, beginning with systems at the government’s digital agency DINUM. The move follows earlier steps including replacing Microsoft Teams with a domestic alternative. Germany‘s digital minister called for developing a European alternative to Palantir, arguing governments should support domestic firms to scale competitive data analytics capabilities.
The CIA has begun using AI to generate its first autonomous intelligence report and plans to expand AI integration across analytic platforms. Deputy Director Ellis said AI will assist with drafting judgments, testing conclusions and identifying trends, while humans retain decision authority. The agency expects AI to become embedded as co-workers within the next few years.
Senior NHS officials warned staff against publicly criticising the rollout of Palantir‘s £330 million Federated Data Platform, amid internal tensions over the system’s adoption. The controversy reflects divisions within the NHS over the role of private tech firms in handling sensitive health data.
The UK government launched a $675 million sovereign AI fund to invest in domestic startups and reduce reliance on foreign technology. OpenAI signed a lease for its first permanent London office with capacity for over 500 staff, reinforcing plans to make the city its largest research hub outside the US, though the company has paused its UK Stargate AI infrastructure project, citing high energy costs and regulatory challenges. The UK government pushed back on OpenAI’s reasoning, with AI Minister Kanishka Narayan suggesting the decision was linked to OpenAI’s internal financial pressures. Microsoft agreed to take over data centre capacity in Norway originally intended for the Stargate initiative, including renting 30,000 Nvidia chips from Nscale.
Maine lawmakers passed a bill freezing construction of new large data centres until November 2027, the first US state to enact such a measure, with the pause applying to facilities of at least 20 megawatts. The bill follows rising public opposition across at least 12 states, though similar measures have largely failed elsewhere. Voters in Festus, Missouri, removed half of the city council over a controversial data centre project, with a lawsuit alleging secret meetings and efforts to suppress opposition. The US Energy Information Administration plans to introduce a mandatory nationwide survey to measure data centre energy consumption — the first federal effort to systematically collect such data.
In Australia, a coalition of Southeast Queensland councils is developing a coordinated strategy to attract data centre investment across Brisbane and the Gold Coast. The sector is facing scrutiny over claims it supports hundreds of thousands of jobs, with analysts saying the widely cited 935,000 figure reflects total tech sector employment rather than data centre-dependent roles, and industry-backed estimates projecting around 17,900 operational data centre jobs by 2030. India is offering a 20-year tax holiday to attract foreign cloud providers, though multiple projects face resistance from farmers and activists over land acquisition and environmental concerns.
Gulf states are continuing large-scale investments in data centres as part of sovereign AI strategies, despite recent drone attacks on facilities in Bahrain and the UAE. Major projects — including a $20 billion partnership between Brookfield Asset Management and the Qatar Investment Authority — remain on track.
The FCC granted Netgear the first conditional exemption from its ban on foreign-made routers, allowing the company to continue selling products in the US despite overseas manufacturing. Other brands remain subject to restrictions, including a 2027 deadline limiting software updates for foreign-made consumer routers.
⸻
💰 Tech business & markets
Meta‘s $2 billion acquisition of Chinese AI startup Manus has drawn scrutiny from Chinese authorities, who launched an investigation and barred two co-founders from leaving the country. British tech investor Fred Blackford has built a stake worth more than $500 million in ByteDance, betting on a valuation gap between the company and Western tech peers. Chinese social platform RedNote is expanding into the US by opening offices, hiring staff and launching an e-commerce portal.
Major US tech firms including Amazon, Meta, Oracle and Block are cutting jobs, with layoffs affecting thousands and slowing overall workforce growth since 2022, according to The Economist, which attributed the contraction to industry adjustment rather than AI-driven displacement. Fintech executive Ben Pfisterer of Zeller argued that recent layoffs are being misattributed to AI, when they are largely driven by over-hiring and organisational inefficiencies following the pandemic.
Half of US employees now report using AI at least occasionally, with daily usage rising, according to Gallup. AI adoption is linked to increased workplace disruption, with both hiring and layoffs more common at AI-adopting organisations, and 23% of employees at such firms expect potential job loss within five years. Bloomberg examined how Gen Z workers are integrating AI tools into their jobs, often accelerating productivity but raising concerns about governance, oversight, and opaque use. Law firms, meanwhile, are facing increased workloads as clients submit large volumes of AI-generated emails and drafts that require extensive review and correction.
AI data-labelling startup Mercor is grappling with internal fraud, suspected North Korean infiltration, and security breaches as it scales to over $1 billion in annualised revenue. An employee embezzled funds via fake contractor payments, while suspected North Korean operatives used stolen identities to contribute training data to major AI labs, with Meta pausing work and others reviewing exposure. Two US citizens were sentenced to a combined 16 years for operating “laptop farms” enabling North Korean IT workers to pose as US-based employees at over 100 companies, generating millions while exposing corporate systems. Separately, crypto platform Drift detailed a months-long social engineering operation by North Korean hackers that led to a $280 million theft.
A report linked a sophisticated Android banking malware campaign targeting Southeast Asian victims to scam compounds in Cambodia, including K99 Triumph City, with operations tied to malware-as-a-service networks and potentially trafficked workers.
Anthropic has introduced identity verification for some Claude users, requiring government-issued ID for cases flagged for potential fraud or abuse. The rollout, handled by third-party provider Persona, has drawn criticism from users concerned about data security.
⸻
🎬 IP, media & creative industries
Voice actors globally are pushing back against the growing use of AI for dubbing and voice generation, with industry groups in countries including Brazil, Mexico and South Korea advocating for legal protections. The shift raises concerns over cultural homogenisation and ownership of voice data as studios adopt AI to reduce costs and scale multilingual content.
A Guardian column described growing concerns about AI-generated content flooding opinion writing, highlighting a new “Proudly Human” certification initiative aimed at verifying human authorship. British tabloids are facing declining print revenues as readers shift to social media, AI-generated summaries, and influencer-led content, with monetisation remaining difficult despite large digital audiences.
Major AI firms including OpenAI and Anthropic are expanding policy engagement through thinktanks, research institutes and lobbying. OpenAI released a policy paper proposing social and economic reforms while opening a Washington office for policymakers; Anthropic launched its own institute. Critics argue these efforts may undercut independent regulation. Anthropic’s $20 million donation to Public First Action cannot be used to influence US federal elections due to restrictions tied to its 501(c)(4) status, with funds designated for public education on AI policy.
⸻
🌏 Global policy
🇦🇺 Australia
An Australian court imposed broad publication restrictions in the espionage case against Igor and Kira Korolev, who are accused of attempting to send classified information to Russia. Prosecutors allege Kira Korolev, an Australian Army information systems technician, accessed defence material via her work account while in Russia, with assistance from her husband in Brisbane. Authorities seized 12 devices and are seeking to protect sensitive evidence including defence capabilities and alliance information.
Australia’s opposition Coalition proposed expanding social media screening for visa applicants and introducing a binding “values” test, with breaches potentially leading to deportation. The plan includes creating an Enhanced Screening Coordination Centre to identify security risks through cross-agency intelligence and monitoring.
🇺🇸 United States
A senior Department of Defense official, Emil Michael, overseeing AI policy sold shares in Elon Musk‘s xAI for between $5 million and $25 million in January, up from a declared value of up to $1 million in 2025. The sale followed Pentagon agreements with xAI, including a December 2025 deal to deploy its AI capabilities. Ethics experts raised concerns about potential conflicts of interest.
President Donald Trump deleted an AI-generated image posted on Truth Social depicting himself as a Jesus-like figure after backlash. The incident intensified tensions within Republican and Catholic voter groups, alongside Trump’s criticism of Pope Leo XIV over US military actions.
The US has shifted from promoting free global data flows to imposing restrictions on cross-border data transfers, driven by national security concerns and competition with China, according to Lawfare. Measures include forcing ByteDance to divest TikTok‘s US operations and limiting sensitive data transfers to foreign adversaries.
Democratic candidates have been advised to avoid antagonising pro-AI groups that have amassed nearly $300 million to influence US midterm elections, with industry-backed Super PACs funding candidates and targeting critics. The Financial Times reported the influx of AI-linked political spending is shaping candidate behaviour ahead of the elections.
Anthropic hosted a two-day summit with Christian leaders in March to seek input on the ethical development of Claude, including how it should respond to grief, self-harm, and complex ethical questions. The effort also surfaced tensions with the US government, which has restricted use of Anthropic’s technology over concerns its built-in ethical constraints could limit military applications.
🇪🇺 Europe
A Politico survey across six EU countries found 84% of Europeans do not trust US tech companies with their personal data, while 93% distrust Chinese firms. Trust in European companies is higher but limited, with only 51% expressing confidence.
European regulators are facing surging workloads as citizens increasingly use AI tools to generate complaints, proposals and submissions to EU institutions. The European Ombudsman reported a 54% rise in complaints in 2025, partly attributed to AI-assisted drafting, straining administrative systems designed before widespread AI use.
🇬🇧 United Kingdom
UK lawmakers were told that China is positioning itself as a supporter of global AI governance, in contrast to a deregulated US approach. Experts warned of national security risks from international AI collaboration and raised concerns about the UK’s reliance on US technology firms, alongside delays in UK AI infrastructure projects.
🇯🇵 Japan
Japanese AI company Fronteo has developed a system to assess the risk of researchers leaking sensitive technologies by analysing funding sources, co-authorship networks, and links to entities in countries of concern. The tool draws on hundreds of millions of academic papers and sanctions lists and aligns with new Japanese government guidelines requiring universities to strengthen research security.
India is expanding partnerships with foreign companies to strengthen its electronics manufacturing sector, with firms such as Japan’s TDK benefiting from government-backed incentives aimed at boosting domestic capabilities.
🇮🇷 Iran
Recorded Future outlined multiple scenarios for the Iran conflict over the next 6–12 months, ranging from a fragile ceasefire with sustained economic disruption to full regional escalation. The baseline scenario sees continued instability, intermittent maritime disruption in the Strait of Hormuz, elevated cyber activity, and sustained pressure on energy markets and supply chains.
Early observations from the Iran conflict, examined in The Australian, note a shift toward low-cost, distributed warfare, with drones, missiles and mobile systems proving more effective than large platforms.
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.



