Trump’s potential Nvidia deal with China raises national security risks

The US President Donald Trump has shattered decades of US national security precedent by striking a deal with Nvidia and AMD that allows the sale of certain banned AI chips to China, but at a certain price.

In an arrangement without modern parallels, the companies will resume exports of their H20 processors to the Chinese market in exchange for giving the US government a 15% share of related revenues.

The move reopens a channel for sensitive technology sales and introduces a transactional element into what had long been treated as a matter of uncompromising national security.

For decades, Washington’s export controls on strategic technologies were blunt instruments: if a product was deemed too sensitive, no amount of corporate lobbying or lost revenue could override the ban.

Trump’s approach breaks from that tradition, effectively monetising access to restricted technologies. He has even floated the idea of allowing a weakened version of Nvidia’s cutting-edge Blackwell chip to be sold in China, a possibility that has set off alarm bells among national security hawks.

Republican and Democratic lawmakers have condemned the decision, warning it risks transforming US security policy into a ‘pay-for-play’ system.

Representative John Moolenaar, who chairs the House Select Committee on China, argued that export controls should remain a first line of defence against adversaries, not a bargaining chip. His Democratic counterpart, Raja Krishnamoorthi, cautioned that putting a dollar value on national security sends the wrong message to both allies and rivals.

The Trump administration has defended the arrangement by downplaying the risk. Commerce Secretary Howard Lutnick called the H20 Nvidia’s ‘fourth-best’ chip, noting that it is already widely used in China. The administration also framed the move to keep Chinese companies tied to US technology rather than turning to rival suppliers. Yet questions loom over the legality of the revenue-sharing scheme.

Trade experts have raised the possibility that it could be interpreted as an export tax, something the US Constitution prohibits, though details of the agreement remain opaque.

Beyond legal debates, the financial implications are significant. Analysts predict the levy could cut gross margins on China-bound chips by as much as 15 percentage points, trimming overall profitability for Nvidia and AMD.

In turn, this change of course could prompt other US companies selling strategic goods to China, from aerospace to advanced materials, to wonder if they too will face similar revenue-sharing requirements.

For some, it could be a costly burden; for others, it might be the only way to retain access to China’s lucrative market.

Would you like to learn more about AI, tech and digital diplomacyIf so, ask our Diplo chatbot!

Trump weighs scaled-down Nvidia chip sales to China

President Donald Trump has signalled that he may permit Nvidia to sell a toned-down version of its latest Blackwell AI chip to China, which could substantially shift US tech export policy.

The idea, still under discussion with Nvidia CEO Jensen Huang, would involve reducing the chip’s computing power by 30% to 50%, creating what Trump described as an ‘unenhanced’ model for the Chinese market. While framed as a compromise, critics warn that even these stripped-down chips could fuel Beijing’s AI ambitions.

The announcement follows an unprecedented agreement between the Trump administration, Nvidia, and AMD, under which the US government would collect 15% of revenue from certain AI chip sales to China.

Washington insiders have expressed unease, noting that, with enough scaled-down hardware, China could still build AI supercomputers capable of competing with or surpassing American capabilities.

Saif Khan, a former White House technology adviser, cautioned that the move could accelerate China’s path toward AI dominance, undoing years of strict export controls.

Currently, Nvidia’s most advanced chip approved for sale in China is the H20, built on older Hopper architecture. The H20 was specifically designed to comply with restrictions imposed under President Biden and entered the Chinese market in 2024.

Although shipments were halted earlier this year, the Trump administration recently granted clearance for exports to resume. Trump dismissed the H20 as ‘obsolete’ and claimed China had already mastered it, suggesting the new Blackwell variant would offer a fresh revenue stream while staying within national security boundaries.

Nvidia’s flagship US Blackwell chip, unveiled in March 2024, is up to 30 times faster than its predecessor, making it a significant leap in AI performance. Details about the proposed Chinese variant remain undisclosed, but Reuters previously reported it would come at a lower cost and reduced power.

The US Commerce Department has begun issuing licenses for the H20, with officials insisting these exports do not threaten national security.

For Nvidia and AMD, the deal represents a rare case of direct government revenue-sharing tied to foreign sales, reflecting Trump’s hands-on approach to corporate negotiations. His administration has previously pressured tech executives to prioritise domestic manufacturing and has intervened in leadership appointments.

Nvidia, for its part, has stated it will follow all US export rules, while AMD confirmed receiving approval to ship some AI processors to China without directly addressing the revenue-sharing clause.

Beijing’s reaction so far has been muted. China’s foreign ministry declined to comment on the potential Blackwell deal but has repeatedly accused Washington of using technology controls to ‘maliciously contain and suppress’ Chinese industry.

Would you like to learn more about AI, tech and digital diplomacyIf so, ask our Diplo chatbot!

Bitcoin jumps after US delays China tariffs deadline

US President Donald Trump has extended the pause on tariffs against Chinese goods by 90 days, postponing the deadline initially set for 12 August. The announcement triggered a sharp rise in Bitcoin’s price, which climbed to nearly $119,800 before returning to around $119,000.

Market watchers remain cautious, debating whether the price surge represents a lasting rally or a brief ‘Dead Cat Bounce.’ A CME futures gap below $119,000 may need to be filled before Bitcoin can sustain an uptrend.

The crypto market also awaits key US inflation data, with consumer and producer price figures expected in mid-August. These statistics could influence the Federal Reserve’s decisions, with a rate cut forecast for September growing more likely.

Crypto analyst Altcoin Sherpa outlined two possible paths for Bitcoin’s near-term movement. One scenario expects a gradual pullback to form a stable base, while the other anticipates a quick liquidity test near $120,000 influenced by treasury and economic factors.

Would you like to learn more about AI, tech and digital diplomacy? If so, ask our Diplo chatbot

AI-powered heist drains $1m from crypto wallets via Firefox add-ons

Hackers have stolen over $1 million in cryptocurrency using AI-generated malicious Firefox extensions disguised as legitimate wallet tools.

The group, known as GreedyBear, created over 150 fake add-ons for platforms like MetaMask and Phantom, bypassing security checks to drain funds from thousands of users. Analysts say AI enabled the attackers to automate coding and deployment at an industrial scale.

The theft comes amid a record-breaking year for crypto crime, with Chainalysis data showing over $2.17 billion stolen so far in 2025. Many incidents exploit smart contract flaws and human error, with access control attacks accounting for the most recent losses.

Security experts warn that AI is now a double-edged sword, helping attackers and defenders. They urge exchanges, developers, and users to adopt AI-powered monitoring, stronger verification, and collaborative defences to restore trust in digital assets.

Would you like to learn more about AI, tech and digital diplomacy? If so, ask our Diplo chatbot!

Google works to curb Gemini’s endless self-criticism

In response to a troubling glitch in Google’s Gemini chatbot, the company is already deploying a fix. Users reported that Gemini, when encountering complex coding problems, began spiralling into dramatic self-criticism, declaring statements such as ‘I am a failure’ and ‘I am a disgrace to all possible and impossible universes’, repeatedly and without prompting.

Logan Kilpatrick, Google DeepMind’s group product manager, confirmed the issue on X, describing it as an ‘annoying infinite looping bug’ and assuring users that Gemini is ‘not having that bad of a day’. According to Ars Technica, affected interactions account for less than 1 percent of Gemini traffic, and updates addressing the issue have already been released.

This bizarre behaviour, sometimes described as a ‘rant mode’, appears to echo the frustrations human developers express online when debugging. Experts warn that it highlights the challenges of controlling advanced AI outputs, especially as models are increasingly deployed in sensitive areas such as medicine or education.

Would you like to learn more about AI, tech and digital diplomacy? If so, ask our Diplo chatbot!

Users warned to update WinRAR after active attacks

A critical flaw in the Windows version of WinRAR is being exploited to install malware that runs automatically at startup. Users are urged to update to version 7.13 immediately, as the software does not update itself.

Tracked as CVE-2025-8088, the vulnerability allows malicious RAR files to place content in protected system folders, including Windows startup locations. Once there, the malware can steal data, install further payloads and maintain persistent access.

ESET researchers linked the attacks to the RomCom hacking group, a Russian-speaking operation known for espionage and ransomware campaigns. The flaw has been used in spear-phishing attacks where victims opened infected archives sent via email.

WinRAR’s July update fixes the cybersecurity issue by blocking extractions outside user-specified folders. Security experts recommend caution with email attachments, antivirus scanning of archives and regular checks of startup folders for suspicious files.

Would you like to learn more about AI, tech and digital diplomacy? If so, ask our Diplo chatbot!

DeepSeek’s efficiency forces OpenAI to rethink closed AI model strategy

OpenAI has released reasoning-focused open-weight models in a strategic response to China’s surging AI ecosystem, led by DeepSeek’s disruptive efficiency. Unlike earlier coverage, the shift is framed not merely as competitive posturing but as a deeper recognition of shifting innovation philosophies.

DeepSeek’s rise stems from maximizing limited resources under the US’s export restrictions, proving that top-tier AI doesn’t require massive chip clusters. The agility has emboldened the open-source AI sector in China, where over 10 labs now rival those in the US, fundamentally reshaping competitive dynamics.

OpenAI’s ‘gpt-oss’ models, which reveal numerical parameters for customization, mark a departure from its traditional closed approach. Industry watchers see this as a hybrid play, retaining proprietary strengths while embracing openness to appeal to global developers.

The implications stretch beyond technology into geopolitics. US export controls may have inadvertently fueled Chinese AI innovation, with DeepSeek’s self-reliant architecture now serving as a proof point for resilience. DeepSeek’s achievement challenges the US’s historically resource-intensive approach to AI.

AI rivalry may spur collaboration or escalate competition. DeepSeek advances models like DeepSeek-MoE, while OpenAI strikes a balance between openness and monetization. Global AI dynamics shift, raising both technological and philosophical stakes.

Would you like to learn more about AI, tech, and digital diplomacy? If so, ask our Diplo chatbot!

Altman warns of harmful AI use after model backlash

OpenAI chief executive Sam Altman has warned that many ChatGPT users are engaging with AI in self-destructive ways. His comments follow backlash over the sudden discontinuation of GPT-4o and other older models, which he admitted was a mistake.

Altman said that users form powerful attachments to specific AI models, and while most can distinguish between reality and fiction, a small minority cannot. He stressed OpenAI’s responsibility to manage the risks for those in mentally fragile states.

Using ChatGPT as a therapist or life coach was not his concern, as many people already benefit from it. Instead, he worried about cases where advice subtly undermines a user’s long-term well-being.

The model removals triggered a huge social-media outcry, with complaints that newer versions offered shorter, less emotionally rich responses. OpenAI has since restored GPT-4o for Plus subscribers, while free users will only have access to GPT-5.

Would you like to learn more about AI, tech and digital diplomacy? If so, ask our Diplo chatbot!

Bo Hines leaves White House crypto role

Bo Hines, executive director of the White House Crypto Council, has announced his departure to return to the private sector. Appointed in December 2024, Hines thanked the crypto community, calling his role ‘the honour of a lifetime’ and pledging ongoing support.

The council, formed to shape US digital asset policy, released a regulatory action plan in July. Despite progress, critics argued it failed to implement a strategic Bitcoin reserve. Deputy director Patrick Witt is expected to succeed Hines, though no official appointment has been made.

Hines strongly backed expanding the government’s Bitcoin holdings through budget-neutral strategies, which is in line with Trump’s January executive order that created a national crypto stockpile.

He previously suggested revaluing US gold reserves, which are priced far below market value. Part of the gains could then be converted into Bitcoin without impacting the federal budget.

Would you like to learn more about AI, tech and digital diplomacy? If so, ask our Diplo chatbot

Beijing 2025 Robot Conference highlights China’s humanoid robotics growth

The 2025 World Robot Conference in Beijing has drawn significant attention to China’s growing humanoid robotics industry. With over 60 humanoid robots on display, the event attracted investors and tech enthusiasts alike, generating a surge in stock prices for companies such as Unitree Robotics.

The conference showcased robots performing diverse activities from industrial operations to more human-like tasks, including marathons and kickboxing, highlighting rapid AI advancements.

China’s Ministry of Industry and Information Technology has supported the sector strongly, aiming for mass production and widespread adoption by 2027. The market, valued at $2.24 billion in 2024, is expected to grow to $41 billion by 2032, reflecting a compound annual growth rate of nearly 44%.

New manufacturing facilities and advances such as carbon fibre materials are boosting the durability and agility of these robots, while companies are focusing on AI integration and teamwork capabilities.

Despite the promising outlook, challenges like high costs, AI learning complexities, and potential overvaluation remain. Experts acknowledge China is closing the gap in humanoid robotics innovation, though technical hurdles persist.

The event underscores the significant role humanoid robots could play in reshaping industries and everyday life, supported by both state initiatives and private investment.

Would you like to learn more about AI, tech and digital diplomacy? If so, ask our Diplo chatbot!