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: 6 June 2026 to 12 June 2026.
Follow the Australian Strategic Policy Institute on Bluesky, LinkedIn, and X.
What We’re Tracking
Tech titans seek to cast opposition to data centers as a Chinese influence operation
What happened: Several recent reports show that China is trying to boost the anti-data center movement in the US — but so far, this research all comes from organisations that are openly aligned with the AI industry or the aggressively pro-AI Trump administration, not from organisations that specialise in China or foreign influence.
Republicans in Congress have called for an investigation into US nonprofits that allegedly receive foreign funding and oppose the data center build-out, but critics say this is a naked attempt to silence dissent and smooth the way for AI industry interests.
OpenAI said it has uncovered what is likely a covert information operation by China-linked actors aimed at boosting popular opposition to the construction of data centers in the U.S. In a report published on June 10, the company said a group of accounts based in China used ChatGPT to generate social media images and text stating that data centers increase electricity costs for American; these were subsequently posted to social media platforms.
Notably, OpenAI said these efforts did not appear to have gotten any traction or moved the needle on US debate.
A report from late May from the Bitcoin Policy Institute, which promotes the crytocurrency, also claimed that China was funding opposition to data centers, as did a similar report from Power the Future, a nonprofit that is aligned with the Trump administration. These two reports were poorly researched, did not show that foreign governments were funneling money to US nonprofits, and made sweeping but unsupported claims that China was the hidden hand behind the anti-data center movement.
Why we’re tracking this: The reports come as local opposition across the US has slowed the mass construction of data centers, which provide the computing power needed to fuel the AI revolution. Data centers consume vast amounts of energy, potentially putting pressure on local power grids and raising electricity costs for local households.
The stakes are high. As the US and China are locked in a tight race for AI dominance, sluggish construction of data centers could give an advantage to the other side. But many Americans worry that data centers are just one more way that the AI revolution is siphoning jobs and money away from struggling Americans and towards an elite few.
What people are saying:
‘It is pretty hard to make the argument this is driven by foreign influence when you are dealing with people in sometimes very small communities showing up at town hall meetings angry about things directly affecting them.’ — Tamara Kneese, senior researcher at the Partnership on AI, The Washington Post
‘The targeting of OpenAI and US data center buildouts is significant not because the operation appears to have shifted public opinion, but because it shows PRC-origin influence operators testing narratives against AI infrastructure.’ — OpenAI
‘Tax-exempt status is not a shield for foreign influence. Organizations that misuse charitable structures to advance foreign interests undermine our laws, our democracy, and public trust.’ — Rep. Jason Smith of Missouri, chair of the House Ways and Means Committee, NOTUS
My view: Foreign government meddling in the domestic debates of democracies is a serious issue. But equally concerning is the age-old trick of trying discredit inconvenient public sentiment by casting it as a foreign ploy.
There is overwhelming evidence that US local opposition to data centers is authentic. There is ample evidence to suggest that the Chinese government does not want the US to build enough data centers. There is no evidence that the Chinese government’s efforts have moved the needle on US debate. It’s that simple.
OpenAI’s research is solid, but the reports from Bitcoin Policy Institute and Power the Future appear to be motivated not by a desire to uncover China’s influence (and thus to preserve the integrity of US civil society), but rather to delegitimise organic movements (and thus to compromise the integrity of US civil society). This is, ironically, a tactic widely used by the Chinese Communist Party to discredit grassroots Chinese civil society by accusing Chinese activists of being in bed with ‘hostile foreign forces.’
This is an anti-democratic abuse of the field of China influence research that both dilutes the impact of true high-quality research, and threatens US civil society itself.
— Bethany Allen, CTS
Anthropic releases Mythos models while calling for mechanism to pause frontier model development
What happened: Anthropic released two models — Claude Fable 5 and Claude Mythos 5 — built on the same Mythos-class system with different safeguards, per WIRED and iTnews. Fable 5 is the first Mythos-class model open to the public.
The release has drawn criticism on two fronts: an undisclosed intervention that, without notifying users, limits the model’s usefulness when it detects use in building rival AI; and a mandatory 30-day data-retention policy overriding existing Zero Data Retention agreements, according to iTnews and Mashable.
Days earlier, Anthropic had published a blog post arguing that its models are increasingly built by AI itself — a path towards recursive self-improvement. The Guardian reported it as Anthropic calling for a worldwide temporary pause on AI development, and for policymakers, researchers, civil society and AI companies to convene on the risks.
The same piece noted the call sits in tension with a Financial Times report that Anthropic embedded engineers in the NSA for offensive cyber operations potentially aimed at Iran and China. Critics questioned whether the proposal reflects safety concern or positioning ahead of Anthropic’s pending $1 trillion IPO filing.
Why we’re tracking this: Several threads in AI development are converging this week. Anthropic is widening access to its most capable model class while arguing the field would benefit from a way to pause development — having held Mythos back for months over its ability to find vulnerabilities in systems such as banking and power grids, per The Guardian. The model’s safeguards are not purely grounded in public-safety concerns: the restriction on using it to build competitor models also serves Anthropic’s commercial position, blurring the line between public-safety and commercial interests.
What people are saying:
‘Our view is we’ve built amazingly powerful technology. We’re going to keep building it.’ — Jack Clark, co-founder of Anthropic
‘Anthropic might give the impression of being warm and fuzzy, but their definition of AI safety is narrow. Supporting US authorities in the development of offensive capabilities has never been something they have spoken against.’ — Steven Murdoch, professor of security engineering at University College London.
‘Next up: Apple randomly reboots your Mac if you’re building competing tech, Gmail silently edits your emails if you mention rival platforms, and Tesla Autopilot swerves if it detects you’re working on self-driving cars. All in the name of safety, of course.’ — Mikel Artetxe, cofounder of Reka AI, an AI lab.
My view: Anthropic’s blog post argued for a mechanism to pause frontier development, not for a pause itself. That is not a new position for the company. In a 2023 interview, Anthropic’s CEO Dario Amodei backed a mechanism that would suspend the development of frontier-models if they were being deployed without adequate safeguards against misuse, the aim being to pressure labs to build those safeguards in advance so the instrument would never had to be triggered. Read that way, the apparent tension between calling for a pause mechanism and releasing its most capable model yet largely dissolves.
The harder question is how such a mechanism would work in practice. To be credible, it would need to include Chinese AI labs alongside their US and allied counterparts. Arms control is the natural governance model to draw on, but frontier AI differs in one respect: the competition is driven primarily by commercial rather than military actors, even as its implications turn strategic.
— Stephan Robin, 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 Department of Defense expanded its 1260H list of Chinese military companies, adding Alibaba, Baidu and BYD and reinstating memory chipmakers ChangXin Memory Technologies and Yangtze Memory Technologies alongside pharmaceutical contractor WuXi AppTec. The designation restricts the firms from US military contracts and research funding and carries few immediate legal penalties, according to Bloomberg.
The US urged NATO allies to use defence budgets to rip out and replace Huawei telecommunications equipment, with State Department China coordinator Joshua Young telling officials in Brussels the allocations could fund the switch; Germany and Spain are leading opposition to wider European Commission proposals to ban high-risk Chinese suppliers over fears of retaliation from Beijing. Taiwan, separately, is negotiating with Washington on much stricter AI hardware export controls that could restrict sales to all customers in China by computing-power thresholds and let Taiwan prosecute Nvidia chip smuggling as a criminal offence for the first time.
A coalition including CXMT, Alibaba, Dongguan Trust, SSCI Leading Fund and Advanced Micro-Fabrication Equipment launched a 3.91 billion yuan (about US$577 million) private equity fund in Shanghai, the Changzhi Hanhai Private Investment Fund, to provide long-term patient capital for China’s hard-tech sectors amid tightening US export controls, according to the South China Morning Post. The move comes as China’s state-backed Big Fund trims stakes in mature chip companies and CXMT prepares for a Star Market IPO.
SpaceX has excluded investors from mainland China and Hong Kong from its initial public offering, and OpenAI is reportedly planning identical restrictions for its own listing later this year, in what The New York Times described as escalating capital and technology decoupling between the United States and China over national security, data governance and intellectual property.
The White House said it would accelerate AI development and use for national security while barring unlawful surveillance and censorship, with President Donald Trump issuing a national security memorandum directing faster AI adoption across intelligence and warfighting domains and requiring updated guidance on autonomous weapons. The memo follows a dispute between Anthropic and the Pentagon over limits on Claude’s use for autonomous weapons and mass surveillance, Reuters reported.
The FBI seized more than a dozen websites it says suspected Chinese agents used to recruit current and former US officials with security clearances, posing as consulting firms advertising defence and ex-military roles and using identity theft and AI-generated photos and videos. The Justice Department separately announced the seizure of 13 domains tied to fake consulting firms that solicited insider information, following a Five Eyes warning that China is increasingly using job platforms to target people.
OpenAI banned two clusters of ChatGPT accounts it said likely originated in China and supported covert influence operations targeting US AI and technology-policy debates, including content claiming AI data centres were raising household electricity prices and material criticising US tariffs. Axios reported the campaigns appeared ineffective but showed pro-China actors testing AI tools to amplify existing US political and economic divisions.
⸻
🧠 AI models, agents & compute
Anthropic launched two Mythos-class models — Claude Fable 5, available publicly with safeguards, and Claude Mythos 5, the same underlying model with some safeguards lifted for trusted users — both priced at US$10 per million input tokens and US$50 per million output tokens, double Claude Opus 4.8 rates. Fable 5 routes cybersecurity, biology, chemistry and model-distillation prompts to Opus 4.8 in fewer than 5% of sessions, while Mythos 5 reached about 200 organisations across more than 15 countries through Project Glasswing. WIRED reported the earlier limited release reflected concern the model could be misused to develop hacking tools. iTnews put Mythos 5 at 10.75 captured Chrome V8 flags on ExploitBench, against 5.56 for Opus 4.8 and 4.44 for OpenAI‘s GPT-5.5.
Anthropic also imposed mandatory 30-day data retention across the Mythos class, overriding existing zero-data-retention agreements with enterprise and API customers — a change copyright and AI lawyer Jessica Eaves Mathews said nullifies prior commitments, which the company justifies as necessary to detect novel attacks and reduce false positives.
Anthropic separately proposed a mechanism to temporarily pause worldwide development of frontier AI models, and proposed convening policymakers, researchers and companies on the risks of advanced systems, citing Claude‘s progress toward recursive self-improvement and reporting that more than 80% of code merged into its codebase in May was authored by Claude. The Economist examined the same recursive-self-improvement concerns and noted Anthropic is expected to list on stock markets later this year. Dario Amodei, Anthropic’s chief executive, argued in an essay that policy is lagging model capabilities and called for binding regulation, including mandatory third-party testing of frontier models for cyber, biological and loss-of-control risks with government power to block unsafe deployments.
In The Australian, Paul Kelly argued Australia’s political and corporate leaders have not kept pace with AI’s economic, social and security implications, contrasting US debate over public stakes in AI companies with warnings from figures including Amodei. A Reuters/Ipsos poll found 53% of Americans fear AI could put them or someone in their household out of work, with 73% worried about increased AI use. Reuters separately reported that Chinese companies are quietly cutting contractors and using attrition as they adopt AI tools while avoiding mass layoffs that could draw government scrutiny, with analysts saying AI adoption is outpacing job creation amid high youth unemployment.
SpaceX struck a roughly US$30 billion deal under which Google will pay about US$920 million a month for AI computing power, giving Google access to around 110,000 Nvidia chips for its Gemini Enterprise agent platform, according to The New York Times. Bloomberg reported Google can terminate the multi-year deal if SpaceX fails to deliver chip access by 30 September 2026 or on 90 days’ notice after the end of the year. Nvidia, separately, said South Korea’s Naver will use its technology to build gigawatt-scale AI factories to meet rising demand for AI and physical-AI applications.
OpenAI is restructuring ChatGPT into an AI superapp combining coding tools and autonomous agents under unified product head Thibault Sottiaux, shifting resources toward its Codex line and enterprise revenue ahead of a planned IPO while shutting down its Sora video tool, the Financial Times reported. OpenAI and Visa also expanded a partnership letting AI agents make online purchases with user permission, mirroring Visa’s parallel deals with Anthropic and Microsoft. DoorDash, meanwhile, announced Ask DoorDash, a chatbot in select markets that lets customers order food and groceries from photos and prompts.
⸻
🛡 Cyber posture
Microsoft removed dozens of GitHub-hosted open-source projects after hackers apparently breached them and injected password-stealing malware, with many affected projects relating to Azure and AI development tools used with Claude Code, Gemini‘s command-line interface and VS Code. Security researchers said the malware could steal credentials when compromised tools were opened in AI coding apps, according to TechCrunch.
ServiceNow warned affected customers that attackers exploited an unauthenticated access flaw through a vulnerable API endpoint to query data from customer instances, and applied a security update earlier this month limiting the endpoint to authenticated users. Administrators linked the incident to the /api/now/related_list_edit/create endpoint and advised reviewing logs, according to BleepingComputer.
France’s digital affairs directorate DINUM said hackers used a compromised account to breach Tchap, the government’s encrypted messaging platform, with ANSSI detecting the breach earlier this week and CNIL alerted. A threat actor claimed responsibility, saying they used social engineering and stole messages, metadata, documents and media files.
Meta fixed a flaw in a customer-service tool that let hackers use an AI-powered chatbot to reset Instagram passwords, affecting roughly 34,000 accounts including 20,000 that were breached and exposed personal data, according to internal documents viewed by The New York Times. Meta said some back-end checks failed but not because of the AI agent itself, and that it was notifying regulators and affected users.
CISA issued a directive requiring US civilian federal agencies to fix, disable or remove the most serious vulnerable software or equipment from the internet within three calendar days, citing hackers’ use of AI and concern that advanced models could enable automated exploitation at scale, Reuters reported. Less severe vulnerabilities will have remediation windows of two weeks to two months.
In the Financial Times, Anthropic was reported to have embedded roughly six forward-deployed engineers inside the NSA to customise its Mythos model for offensive cyber operations such as infiltrating foreign networks. The arrangement persists despite Anthropic’s lawsuit against the Department of Defense over a supply-chain risk designation linked to the company’s attempts to restrict Claude’s use in lethal drones and mass surveillance.
⸻
🕵️ Surveillance states
German cybersecurity journalist Marc Hofer uncovered an unsecured prototype Chinese policing dashboard titled Dynamic Control Platform for Overseas Personnel, apparently built for Zhangjiakou‘s Public Security Bureau, containing profiles of foreign journalists and residents with passport details, phone numbers, CCTV records, travel data, hotel stays and relationship-mapping functions, according to The Sydney Morning Herald. The platform was closed off within hours of Hofer and former Telegraph correspondent Sophia Yan publishing their findings.
Meta said it is filing a federal-court contempt motion against Israeli spyware firm NSO Group for allegedly violating an injunction barring it from targeting WhatsApp users, after WhatsApp disrupted new spear-phishing attempts resembling earlier one-click campaigns. Civil-rights groups, security researchers and digital-rights experts filed amicus briefs opposing NSO’s appeal, Reuters reported.
Meta deleted code for an unreleased face-recognition system, internally called NameTag, from its Meta AI smart-glasses companion app after WIRED reported on the feature, which was designed to build biometric faceprints from faces captured by the glasses and match them against a local database. Meta said the feature was exploratory and no final decision had been made.
Russia temporarily disabled a specialised surveillance network protecting President Vladimir Putin and top aides after the assassination of Iran’s Supreme Leader Ayatollah Ali Khamenei, the Financial Times reported. The precautions followed revelations that Israeli intelligence weaponised advanced language-based AI to query millions of hours of Iranian CCTV data to track high-value targets.
Microsoft said it will tighten human-rights controls when working with national-security agencies after an inquiry into Israeli military use of its cloud for mass surveillance of Palestinians, which found Unit 8200 had used Azure to store intercepted phone calls. Microsoft terminated the unit’s access to some cloud and AI services and said it would adopt the inquiry’s recommendations, according to The Guardian.
The US House rejected a stopgap measure to renew Section 702 of the Foreign Intelligence Surveillance Act, all but ensuring the warrantless-wiretapping authority lapses this week after renewal talks collapsed over the status of acting intelligence chief Bill Pulte. Legal experts note a March surveillance-court recertification will likely require telecommunications companies to keep complying with intelligence directives into 2027, even as reformers push for warrant requirements and limits on purchased broker data.
⸻
⚖️ Platform accountability
Meta plans to use data shared by other businesses to personalise Facebook and Instagram feeds and AI responses, expanding an approach it already applies to ads, according to The Verge. Meta said the change does not involve collecting new data and can be controlled through its Activity from other businesses setting.
Chinese activist Apple Peiqing Ni was targeted on X with deepfake posts portraying her as promiscuous and a drug user after she posted about attending a Tiananmen massacre commemoration; X initially told her the posts did not breach its harassment or violent-speech rules, then suspended the account after The Guardian contacted its press office. Ni said UK police had advised her to report the posts to X and that her parents in China had been harassed over her activism.
⸻
🧑⚖️ Courts, enforcement & regulation
A federal judge in Mississippi sanctioned lawyers on both sides of a contractual dispute after finding they filed AI-generated briefs citing nonexistent cases, with Senior US District Judge Sharion Aycock cancelling the trial, disqualifying all four lawyers, barring two from the court for two years and fining each between US$1,000 and US$3,500, according to 404 Media.
A Munich regional court issued a temporary injunction against Google over false claims in AI-generated search overviews about two publishers, treating the overviews as Google’s own content and making it directly liable rather than applying traditional search-engine protections, The Decoder reported. The court rejected Google’s argument that users could verify linked sources themselves.
⸻
🏛️ Government, procurement & public sector tech
Britain is reviewing its £330 million NHS contract with Palantir amid pressure to trigger a break clause when the initial term ends in early 2027, with technology minister Liz Kendall saying the review will weigh patient confidentiality, public trust and reliance on a US supplier. A parliamentary committee urged ministers to end the contract, warning Palantir’s role posed an unacceptable point of weakness, according to Reuters.
Australia’s Department of Parliamentary Services announced the Parliamentary Information and Cyber Resilience project, the most significant upgrade to the federal Parliamentary Computer Network since its creation, aiming to segment the network and close compliance gaps with the Essential Eight standards. The move follows an Australian National Audit Office report that found the department only partly effective, citing failure to properly implement seven of the Australian Signals Directorate‘s Essential Eight strategies and a 55% turnover rate in its Cyber Security Branch over a year.
⸻
🏗️ Data centres & energy
In Crikey, Ketan Joshi wrote that a Greenpeace Australia Pacific report he led calls for a nationwide moratorium on data centre development in Australia, arguing the technology industry will not produce enough renewable energy to match new demand and that gas-fired data centres could worsen climate impacts. Joshi said Australia should pause, regulate and restrict data centre development.
Texas Governor Greg Abbott released recommendations for state data centre regulation ahead of the 2027 legislative session, proposing that new facilities add power generation, pay grid-interconnection and infrastructure costs, use closed-loop water systems, report electricity and water use and lose sales-tax exemptions. Abbott also directed regulators to act sooner on transmission and infrastructure costs.
NEXTDC‘s M3 facility in Melbourne — described by Crikey as a 225-megawatt site and one of Australia’s largest data centres — is promoted as a new digital gateway for the state, with 41,000 square metres of technical space built for hyperscale growth. The article framed M3 as part of the rapid expansion of data centres on industrial estates around Australian capital cities.
⸻
💰 Tech business & markets
President Donald Trump signalled openness to the US government taking equity stakes in leading AI labs through a public-private structure seeding a government wealth fund, Bloomberg reported, following proposals from OpenAI‘s Sam Altman and a push by Senator Bernie Sanders to require AI companies to surrender half their stock to redistribute wealth to citizens facing job displacement. Critics argued direct federal ownership risks nationalising the sector.
In The New York Times, former Andreessen Horowitz general partner John O’Farrell argued that major AI investors are using political spending to deter regulation, citing the Leading the Future PAC backed by AI and tech figures and the pro-regulation Public First Action PAC backed by Anthropic executives. O’Farrell said large-scale AI campaign spending distorts democracy and should instead fund public education and AI public-interest projects.
In Axios, Elon Musk was described as having built a corporate empire investors treat as too important to penalise despite repeated controversies, with the piece noting he was using X for far-right culture-war commentary on the eve of SpaceX‘s expected IPO. It framed investor tolerance of Musk’s conduct as unusually broad compared with standards applied to other chief executives.
Kalshi plans to require users in some prediction markets to disclose their employers as a guardrail against insider trading and manipulation, applying the rule to markets tied to material nonpublic information such as company performance and national security, according to The Wall Street Journal. The change follows an audit-committee recommendation and rising scrutiny from lawmakers, regulators and prosecutors.
Wesfarmers chief executive Rob Scott said the group will use AI to lift productivity across its stores and withstand weak consumer conditions, targeting existing pain points and freeing staff from repetitive tasks, according to the Australian Financial Review. Scott said Bunnings and Kmart marketplaces were growing after Wesfarmers closed Catch.
Jedify, a New York-based startup, raised US$24 million in Series A funding led by Norwest, with participation from Snowflake and others, to build a context graph connecting enterprise knowledge sources so AI agents can use company-specific data, permissions and workflows. The company said it has 10 to 20 early customers and is targeting mid-market and large enterprises.
⸻
🧒 Online harms & child safety
Apple announced child-safety changes at WWDC 2026 expanding parental controls across its devices, including broader Ask permissions for websites and unknown contacts, automatic filtering of potentially inappropriate images, FaceTime protections and more detailed Screen Time settings, according to Engadget. Apple said it is collaborating with the American Academy of Pediatrics on healthy screen-time guidelines.
Apple chief executive Tim Cook briefed Prime Minister Anthony Albanese on the new controls and, according to Albanese’s office, credited Australia‘s under-16 social-media ban as part of the inspiration, with tools including child-account setup changes, Ask to Browse, app-category time limits and expanded Communication Safety protections. The Sydney Morning Herald contrasted Apple’s device-level controls with Australia’s platform-level ban, which eSafety commissioner Julie Inman Grant has questioned.
UK Prime Minister Keir Starmer said big tech firms operating in Britain must stop children circulating nude images or face legislation, wanting Google, Apple and others to build or activate device controls to detect and block such images for children while allowing adult access through age verification, Reuters reported. Starmer is also reportedly considering restrictions on harmful social-media platforms for under-16s.
⸻
🌏 Global policy
🇺🇸 United States
The White House is negotiating federal preemption of some state AI laws in exchange for support for congressional tech-policy priorities, Axios reported, as states pass stronger AI rules. Representatives Jay Obernolte and Lori Trahan separately unveiled a 269-page bipartisan discussion draft that would preempt some state AI laws while requiring leading developers to disclose safety and security risks, use third-party auditors and formally establish the Center for AI Standards and Innovation.
Senators Ted Cruz and Ron Wyden introduced the bipartisan JAWBONE Act to curb federal pressure on broadcasters, online services and AI services to restrict speech, letting people sue federal agencies or employees for coercive jawboning and allowing state attorneys-general to bring civil actions, according to Ars Technica. It would also require certain government communications with social-media, AI and broadcasting companies to be logged to a public-summary portal.
A federal judge vacated President Donald Trump‘s policy imposing a US$100,000 fee on employers’ H-1B visa applications, with Judge Leo Sorokin ruling it violated the Administrative Procedure Act and the Constitution as a tax Congress had not delegated to the executive, CNBC reported. The administration said it plans to appeal.
🇪🇺 Europe
EU regulators criticised Apple for blaming the Digital Markets Act for its decision not to roll out upgraded Siri AI in the bloc for now, after the European Commission rejected Apple’s request for an 18-month exemption from interoperability obligations. The Commission said the decision was Apple’s alone and that the DMA does not block new product launches, Reuters reported.
The European Commission ordered Meta to give rival AI chatbots free access to WhatsApp while it investigates whether Meta abused its market power, issuing the interim measure after complaints from AI-assistant developers about Meta’s access fees. Meta criticised the order as regulatory overreach and said it would appeal.
EU Drugs Agency executive director Lorraine Nolan warned that drug gangs are exploiting AI, drones and online chemical marketplaces to create new designer chemicals and evade detection, with Europe emerging as a synthetic-drug production hub, the Financial Times reported. The agency’s annual report said new psychoactive substances were detected at roughly one a week last year and seizures hit a record by quantity in 2024.
🇬🇧 United Kingdom
The UK government introduced the National Security (State Threats) Bill to Parliament, creating counter-terrorism-style powers against foreign state organisations and state-linked proxy groups, allowing the Home Secretary to designate hostile organisations and creating offences for support or payment linked to them. The Home Office published accompanying documents, noting the bill builds on the National Security Act 2023 and applies new criminal sanctions to designated bodies, including proxies used by states to carry out hostile activity in the UK.
🇨🇦 Canada
Canada unveiled an AI for all strategy it says will create 250,000 jobs by 2031 and lift GDP by 3%, with Prime Minister Mark Carney announcing a C$500 million Canadian Tech Growth Fund and financing to help smaller businesses access AI tools, Reuters reported. Canada separately introduced a digital-safety bill that would ban social media for children under 16 unless platforms meet safety standards and create a regulator to set standards for AI chatbots, with penalties up to 3% of global revenue or C$10 million.
🇮🇳 India
India has effectively frozen final approvals for Starlink to begin commercial operations, with security agencies under the Ministry of Home Affairs withholding clearances over concerns about satellite-terminal use during the Iran war and seeking assurances despite Starlink’s US ownership, Bloomberg reported. The delay has also stalled a satellite-spectrum pricing proposal needed for commercial launches by Starlink and Indian competitors.
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.
Both lead items share a shape the digest handles separately: a genuine public concern used as the wrapper for a commercial one. Foreign influence is real, but it does convenient work for an industry that needs the build-out unblocked. AI safety is real, but a safeguard that degrades the model when it catches you building a rival is hard to read as anything but moat maintenance. With over $725 billion in hyperscaler capex committed, the people showing up angry at town halls aren't thinking about Beijing. They're asking whose electricity bill funds the gains, and who captures them.
import time
import os
import psutil
import math
def lock_to_core():
p = psutil.Process(os.getpid())
p.cpu_affinity([0])
# Set to high priority to kick other processes off our core
try: p.nice(-20)
except: p.nice(psutil.HIGH_PRIORITY_CLASS)
def get_silicon_truth(chunk_size):
total_iterations = 20000
t0 = time.perf_counter_ns()
chunks = total_iterations // chunk_size
for _ in range(chunks):
for in range(chunksize):
_ = math.sqrt(3.14159) / 1.618
for in range(totaliterations % chunk_size):
_ = math.sqrt(3.14159) / 1.618
return time.perf_counter_ns() - t0
def run_the_gauntlet():
lock_to_core()
print("--- THE GAUNTLET: THE FINAL SETTLEMENT ---")
print("TESTING: 64 (CS Aligned) vs 110 (Harmonic) vs 111 (Non-Harmonic Control)")
# Track the best jitter for each to see their "Peak Stability"
records = {64: [], 110: [], 111: []}
# 50 Rounds of pure stress
for i in range(50):
for chunk in [64, 110, 111]:
samples = [get_silicon_truth(chunk) for _ in range(10)]
jitter = max(samples) - min(samples)
records[chunk].append(jitter)
if i % 10 == 0:
print(f"Progress: Round {i}/50...")
print("\n--- FINAL VERDICT ---")
print(f"{'CHUNK':<10} | {'MIN JITTER':<15} | {'AVG JITTER':<15}")
print("-" * 45)
for chunk in [64, 110, 111]:
m_jitter = min(records[chunk])
a_jitter = sum(records[chunk]) / len(records[chunk])
print(f"{chunk:<10} | {m_jitter:<15.1f} | {a_jitter:<15.1f}")
if name == "__main__":
run_the_gauntlet()