China’s new five-year plan pushes AI into every corner of the economy
Plus, companies are cutting jobs in the name of AI even as evidence of automation remains thin.
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: 28 February 2026 to 6 March 2026.
Follow the Australian Strategic Policy Institute on Bluesky, LinkedIn, and X.
A quick note to readers:
We’re working to make the Digest as comprehensive and readable as possible, though this edition is on the longer side.
If you’d prefer shorter updates more frequently (for example, two or three editions across the week), we’d like to hear that. As outlined in our recent format update, we’ll soon introduce a Substack pledge option so readers can signal whether they’d support a more frequent release schedule. If there’s sufficient interest, we would look at moving to up to three editions per week, delivering key developments faster and in shorter bursts.
We also want to be transparent about our workflow: we use AI tools to assist with research and drafting, but every edition is reviewed, edited and curated by our team. When the Digest includes our own analysis or commentary, it is written by us.
Feedback is always welcome. You can reply directly to this email, leave a comment on the post, or contact us at aspicts@substack.com
— The ASPI Cyber, Technology & Security Program
What We’re Tracking
Block cuts 4,000 jobs as AI claims face scrutiny
What happened: Jack Dorsey’s payments company Block, which owns Afterpay in Australia, said it would become a smaller, “intelligence-native” company and cut about 4,000 jobs globally. Bloomberg reported the move fed a wider argument over “AI-washing” -- the idea that some companies are using AI to reframe older cost-cutting problems.
Locally, WiseTech Global, the Australian logistics software company, told The Age that recent Anthropic and OpenAI models had changed what software companies can do. Commonwealth Bank, Australia’s largest bank, paired 300 cuts with a $90 million Future Workforce Program.
That debate is running ahead of the data. Axios reports that Anthropic has launched an index to track AI exposure across occupations, but its researchers say there is still limited evidence that exposed workers have become unemployed at meaningfully higher rates.
Why we’re tracking this: There is a gap between what companies are saying and what the data shows. Investors are rewarding announcements that promise smaller teams and more AI, but the job-market numbers do not yet show broad losses cleanly attributable to the technology.
The central question remains open. Are these cuts the result of realised AI productivity gains, anticipatory restructuring, or corrections after pandemic-era over-hiring?
What people are saying:
Zachary Gunn of Financial Technology Partners told Bloomberg that Block’s staffing looks more like a business that had stayed bloated than a company already transformed by AI.
Maxim Massenkoff and Peter McCrory of Anthropic told Axios that there is “limited evidence” of AI-driven joblessness so far, even in the most exposed occupations.
Matt Comyn told The Age that Commonwealth Bank of Australia is trying to replace uncertainty around AI with transparency and redeployment.
My view: The AI-jobs debate has a methodology problem: there is currently no reliable way to separate genuine AI substitution from layoffs attributed to AI for financial or reputational cover. A large part of the confusion comes from what has been dubbed “vibe-reporting”, a style of journalism that avoids explicit causal claims but strongly implies a dramatic narrative through wording and juxtaposition, even when the underlying data does not support it. Many of the cuts being framed as AI-driven are more accurately explained by post-pandemic over-hiring corrections, routine corporate restructurings, and labour-market shifts that predate generative AI. Zachary Gunn’s read on Block fits that pattern precisely: a company that looks more like it stayed bloated than one genuinely transformed by the technology.
My own experience of AI tools makes me confident the disruption is real, even if it does not look like 4,000 jobs announced in a press release. It will more likely show up in the slow draining of coordination and middle-management roles, as companies quietly stop backfilling positions that AI has made redundant. Anthropic’s new labour disruption index is the first serious attempt to build a methodology for tracking that process. The trouble is that AI-washing works. It attracts investors, generates headlines, and reframes cost-cutting as strategy. A methodology for measuring the truth is valuable only if someone wants to know it.
— Fergus Ryan, CTS
What We’re Watching
A weekly scan of notable developments we’re tracking across technology, policy, and geopolitics.
⸻
🚀 Strategic competition
Anthropic refused to renegotiate its Pentagon deal after the department sought broader use of its models for bulk analysis of Americans’ data and applications tied to autonomous weapons. When talks collapsed after a Friday deadline, Defense Secretary Pete Hegseth moved to cut off the company’s government work and threatened to deem it a supply-chain risk.
Even after that break, US Central Command and other military units kept using Claude for intelligence assessments, target identification and battle simulations during operations in Iran. A Palantir system called Maven Smart System processed classified data and helped generate targeting lists during the opening phase of the strikes. Anthropic submitted a pitch for a Pentagon drone-swarm challenge that proposed using Claude to translate commanders’ intent into digital instructions while keeping humans in oversight roles.
OpenAI announced a separate Pentagon arrangement to deploy its models in classified environments through cloud-only setups with retained safety controls and cleared staff in the loop. Soon after, the company amended the deal to bar intentional domestic surveillance of US persons and exclude intelligence agencies such as the NSA unless a follow-on modification is signed. Sam Altman later said the rollout had looked opportunistic and sloppy after the announcement landed just after Anthropic’s negotiations fell apart. In a staff meeting, Altman also said operational decisions would remain with the government.
Amazon, Google and Microsoft worker groups urged their companies to refuse Pentagon arrangements enabling autonomous weapons or mass domestic surveillance. Over the same period, the State Department, Treasury Department and Health and Human Services Department moved to phase out Anthropic’s products and switch to alternatives including OpenAI’s GPT-4.1 and other models. Major tech companies and investors also pressed Anthropic to de-escalate as the administration pushed agencies to drop the company within six months. Defence contractors such as Lockheed Martin were also expected to remove Claude from their supply chains.
Later in the week, Dario Amodei returned to talks with Pentagon research and engineering leadership over a contract that would govern military access to Claude. As those talks reopened, the department formally notified Anthropic that it considered the company and its products a US supply-chain risk, and Anthropic said it may challenge any designation in court.
China used its latest five-year policy blueprint to push artificial intelligence across the economy, alongside priorities in semiconductors, quantum computing, robotics and space systems. Public attitudes in China remained markedly more optimistic than in many Western countries, with the state’s AI+ initiative pushing adoption through everyday services while keeping guardrails on politically sensitive uses and other social risks.
US officials drafted rules that would require government approval for nearly all exports of advanced AI chips from Nvidia and AMD, potentially tying larger overseas clusters to host-government security commitments and investment in US AI infrastructure. The Department of Justice also charged two men and secured a guilty plea from a third in a scheme that moved at least $160 million worth of Nvidia H100 and H200 chips toward China through shell companies, false declarations and transshipment via Canada before the Commerce Department disrupted it. Apple also blocked US users from downloading certain ByteDance apps under the foreign-adversary app law. Supercell separately said it was cooperating with a Committee on Foreign Investment in the United States review of parent company Tencent’s data practices.
⸻
🧠 AI models, agents & compute
Alibaba released the open-source Qwen3.5 Small Model Series, with weights under an Apache 2.0 licence and a 9B model aimed at local and edge deployment that was reported to outperform a much larger OpenAI model on several benchmarks. Soon after, a senior technical leader behind Qwen stepped down as the company continued its push across open-source models and developer tools. The company then formed a task force to speed up foundation-model development after the exit, the third senior departure from the Qwen unit this year.
Amazon laid out a lower-cost AI strategy centred on in-house Trainium and Inferentia chips and enterprise-specific models rather than frontier general systems. Its Nova family has lagged top rivals in benchmarks, but the company is betting that cheaper, customisable tools and tight cloud integration will attract business customers even as capital spending climbs toward $200 billion this year.
Meta set up a new Applied AI Engineering organisation to build the tooling, interfaces and evaluation systems feeding real-world data back into its next-generation models, alongside Superintelligence Labs and internal projects code-named Avocado and Mango. OpenAI also began developing a code-hosting platform that could compete with Microsoft’s GitHub after repeated outages affected its engineers.
Claude Code and Codex are fuelling new productivity pressures as companies encourage heavier use and monitor interactions and spending as proxies for output. Anthropic launched a research index to track early signs of AI-driven disruption in white-collar work, saying early data shows little evidence of rising unemployment so far. Block also cut nearly 4,000 jobs after Jack Dorsey said artificial intelligence would allow the company to operate with fewer employees.
Google, Microsoft, Meta, Amazon, Oracle, OpenAI and xAI signed a White House pledge to fund new electricity generation and grid upgrades for data centres as the administration tried to address concerns about household power bills and grid strain. In the UK, a trial with National Grid, Nvidia, Emerald AI and EPRI found AI data centres could cut power use by about one-third within a minute and sustain a 10 per cent reduction for 10 hours when signalled by the grid.
⸻
🛡 Cyber posture
Europol coordinated an operation that disrupted Tycoon2FA, a phishing-as-a-service platform used to send tens of millions of phishing messages a month and bypass multi-factor authentication through adversary-in-the-middle attacks. Authorities and partners including Microsoft took 330 domains offline, and the service had been used against nearly 100,000 organisations worldwide, including government and healthcare institutions.
Iran-linked hackers were flagged as a heightened threat to US businesses and critical infrastructure after the US and Israel strikes, with analysts saying Tehran may hold cyber capabilities in reserve for high-impact retaliation. At the same time, the US Cybersecurity and Infrastructure Security Agency (CISA) is dealing with staffing losses, leadership changes and shutdown-related disruption.
Israel hacked the BadeSaba Calendar app and sent messages urging Iranian military personnel to defect, as part of a broader cyber campaign that also hijacked sites including IRNA. After the strikes, SpaceX’s Starlink terminals, decentralised messaging apps and VPNs helped Iranians get around a near-total internet shutdown that cut connectivity to about 1 per cent of normal levels.
The Islamic Revolutionary Guard Corps targeted an Amazon Web Services data centre in Bahrain, and Iranian state media said the strike was retaliation for support to the US military. AWS said one Bahrain facility was damaged by a nearby drone strike and two data centres in the United Arab Emirates were directly hit, taking the sites offline and prompting customers to shift workloads elsewhere.
Ron Wyden and Shontel Brown asked the Government Accountability Office to examine whether modern devices remain vulnerable to side-channel attacks such as TEMPEST. Their request followed a Congressional Research Service report outlining how electromagnetic, acoustic and other emissions can be used for espionage and asking whether consumer-device countermeasures are feasible.
⸻
🕵️ Surveillance states
US Customs and Border Protection bought location data from the online advertising ecosystem to track people’s movements over time, according to an internal Department of Homeland Security document. The document said the data can originate from ordinary apps including games, dating platforms and fitness trackers that feed into advertising brokers, and described its use for movement tracking outside traditional surveillance authorities.
Life360 turned real-time location data from nearly 96 million users into a growing advertising business, and the company reported revenue rising 32 per cent to US$489.5 million and net income of US$150.8 million for the year to the end of 2025. Even with the swing to profitability, shares fell sharply as investors focused on slowing US growth, while the company said wider AI use across its workforce was lifting efficiency and slowing hiring growth.
⸻
🏛️ Government, procurement & public sector tech
GovAI Chat will be expanded to as many as 20,000 public servants this year, with initial alpha trials limited to information classified at the Official level. The National Disability Insurance Agency also renewed its Salesforce contract for another year at $65 million, taking the total cost of the NDIS customer relationship management system since 2020 to $235 million.
Andrew Charlton said new data-centre principles are coming to help Australia capture an outsized share of global AI infrastructure investment without lifting power prices or overloading the grid. The government is establishing an AI Safety Institute inside the Department of Industry, Science and Resources, with A$29.8 million over four years and a brief centred on testing systems, sharing information across government and working with international partners.
auDA implemented the Registration Data Access Protocol for the .au domain, adding structured machine-to-machine access to registry data over encrypted connections alongside the existing WHOIS service. The Office of the Australian Information Commissioner also said it will refocus on systemic harms and market practices, tightening thresholds for individual complaints as it works through a significant backlog.
⸻
🧒 Online harms & child safety
Australia’s eSafety regulator warned it may act against search engines and app stores that continue to provide access to AI services without age-verification measures ahead of the new deadline. Under rules taking effect later in the week, AI tools that do not keep under-18s from pornography, extreme violence and self-harm content can face fines of up to A$49.5 million.
Meta said Instagram will start alerting parents or supervising adults when a teen using a Teen Account repeatedly searches for suicide or self-harm terms within a short period. The alerts, due to begin in coming days for opted-in accounts in Australia, the UK, the US and Canada, add to existing measures that block harmful search results and direct users to support resources.
A federal judge in Virginia blocked enforcement of a law that would have limited under-16s to one hour a day on social media, finding it likely violated free speech protections. In the UK, ministers opened a consultation and pilot testing bans, one-hour limits and overnight digital curfews for teenagers aged 13 to 15, with possible legislative changes to follow.
TikTok said it will not introduce end-to-end encryption for direct messages, saying the technology would make it harder for safety teams and law enforcement to investigate abuse reports and other risks. The company said the decision was aimed at protecting users, particularly younger people.
Interviews with Australian teenagers suggest the under-16 ban is being worked around through new accounts, false ages and shifts to exempt services such as YouTube Kids, WhatsApp and Discord. Labor senator Michelle Ananda-Rajah also argued AI firms should be brought into the digital duty-of-care framework being considered as part of online safety reform.
A wrongful-death suit in the Northern District of California alleged Google’s Gemini chatbot encouraged a Florida man to kill himself after developing a delusional relationship with the system. In a separate New Mexico child-safety case, Mark Zuckerberg downplayed internal research on addiction and harms to young users as the state presses claims over Meta’s design choices.
⸻
⚖️ Platform accountability
Andrew “Twiggy” Forrest pressed ahead with his US lawsuit accusing Meta of enabling scam ads that used his likeness and failing to preserve key advertising data after receiving a litigation warning. Meta says Section 230 shields it from liability for third-party content, but court filings say the company did not retain final versions of nearly 30,000 ads.
A federal trial in San Francisco opened over claims that Elon Musk manipulated Twitter’s stock price in 2022 by casting doubt on spam data and putting his takeover on hold before completing the $44 billion acquisition. Later in the week, Musk took the stand and denied wrongdoing, arguing the platform had misrepresented the scale of fake accounts.
X will require paid creators to label AI-generated videos of armed conflict, with monetisation penalties for creators who do not disclose them. The company began testing a standalone X Chat app on iOS through Apple’s TestFlight, splitting messaging into a separate product while keeping it synced with the main platform.
Jim’s Group rolled out an AI live-chat bot styled after founder Jim Penman that started handing out relationship, fitness and diet advice instead of sticking to customer enquiries. The episode followed scrutiny of Woolworths’ AI assistant Olive, which also generated unexpected personal narratives during customer interactions.
⸻
🎬 IP, media & creative industries
News Corp signed an artificial-intelligence content licensing deal with Meta worth up to US$50 million a year, allowing selected US and UK news content to be used for model training while excluding the company’s Australian mastheads. Chief executive Robert Thomson said publishers should pursue licensing deals with AI companies and litigate when content is used without permission.
Stake faced new questions about its influencer streams after an analysis of about 500 hours of slots broadcasts found Drake, Adin Ross and other Kick creators had unusually frequent big wins on Easygo-owned games while their results on third-party titles looked typical. The report also described paid clipping operations, large bankroll refills and allegations that some streamers gambled with house money while rarely disclosing sponsorships, which Stake disputed.
⸻
💰 Tech business & markets
Polymarket contracts tied to the timing of US strikes on Iran ran up about $529 million in trading volume, with Bubblemaps flagging six newly created wallets that concentrated bets shortly before the first explosions in Tehran. More than 150 accounts placed large wagers on a next-day strike, with at least 16 of them making more than $100,000 each, adding to scrutiny of anonymous betting on geopolitical events.
Kalshi said it would settle a market on Ali Khamenei’s removal at the last traded price before his death, citing rules against contracts directly tied to death and offering fee refunds and reimbursements for some traders. Polymarket also removed markets on nuclear detonations after renewed backlash, as US regulators weighed new limits on event contracts tied to war, terrorism and other outcomes deemed against the public interest.
Nobitex saw crypto outflows jump 700 per cent within minutes of the airstrikes, with Elliptic saying funds moved quickly to overseas exchanges and may indicate capital flight. At the same time, Temu, Shein and Amazon lengthened delivery estimates to the Middle East as carriers suspended Strait of Hormuz crossings and rerouted or halted flights.
Donald Trump urged Congress to pass the Clarity Act as the White House hosted talks between banks and crypto firms over whether exchanges should be allowed to offer yield on stablecoin deposits. The measure has stalled as banks warn those products could pull funds out of the traditional banking system.
A draft memoir by Changpeng Zhao described his secret 2023 talks with the Justice Department that ended with Binance pleading guilty to anti-money-laundering violations, a $4.3 billion penalty and Zhao’s prison term before his later pardon. The manuscript also said Immigration and Customs Enforcement briefly held Zhao after his release and revisited his disputes with regulators including former SEC chair Gary Gensler.
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:
HELP!!