عربي English Türkçe اردو فارسى
Logo
Latest News
Technology

China Says US Accusations of AI Theft Are ‘Groundless’

The logo of DeepSeek is seen during the Global Developer Conference, organized by the Shanghai AI Industry Association in Shanghai on February 21, 2025. (AFP)
The logo of DeepSeek is seen during the Global Developer Conference, organized by the Shanghai AI Industry Association in Shanghai on February 21, 2025. (AFP)
TT
TT

China Says US Accusations of AI Theft Are ‘Groundless’

The logo of DeepSeek is seen during the Global Developer Conference, organized by the Shanghai AI Industry Association in Shanghai on February 21, 2025. (AFP)
The logo of DeepSeek is seen during the Global Developer Conference, organized by the Shanghai AI Industry Association in Shanghai on February 21, 2025. (AFP)

China said on Wednesday that US accusations of Chinese labs stealing the capabilities of American models were "groundless", as the two countries race for tech supremacy. 

The US cyber defense agency released an advisory on Tuesday accusing top Chinese AI labs including DeepSeek and Moonshot AI of systematically stealing the capabilities "through industrial-scale knowledge distillation campaigns". 

It came ahead of an expected meeting this month between President Donald Trump and his Chinese counterpart Xi Jinping at the White House, where AI could be an important discussion topic. 

China hit back on Wednesday, calling the accusations "groundless". 

"The US approach is a typical case of using the crackdown on distillation as a pretext to achieve industrial monopoly," the Ministry of Commerce said in a statement. 

The US accusations are a "case of double standards", it added. 

Distillation is where an AI model is trained to mimic the output of a larger, more capable one. 

The method is widely used across the industry, but Washington says Chinese firms have deployed it at scale to illicitly copy American systems. 

The US advisory also detailed how DeepSeek, Moonshot AI, Alibaba, MiniMax, StepFun and Z.AI allegedly used millions of exchanges and requests to steal the capabilities of US models from the likes of OpenAI, Anthropic, Google and xAI. 

Chinese companies that adopt this strategy "see significantly shorter AI development timelines and reduced financial expenditures in training a frontier model", the advisory said. 

These actions, taken "likely with Chinese government awareness", have occurred since at least late 2024, it said. 

AFP contacted the Chinese companies apart from StepFun, which was not immediately reachable. Moonshot AI declined to comment. 

Trump is under pressure to respond to the increasing success of Chinese AI models that many companies prefer to the more powerful, but pricier, US versions from Anthropic or OpenAI. 

In July, dozens of startups wrote a letter to the Trump administration urging it to think twice against any outright ban or strict curbs on foreign-made AI models, saying this should only be considered for government use of the technology. 

That month, Beijing also turned the tables by accusing US AI of using Chinese examples to train their models. 

Asked about the accusations at a regular briefing on Wednesday, Chinese foreign ministry spokeswoman Mao Ning said the United States should "refrain from making unfounded accusations against and smearing China". 

"The development of artificial intelligence in China is the result of achieving high-level self-reliance and strength in science and technology," she said. 

"Both China and the United States are major countries in the field of AI. We should strengthen cooperation." 

Most Viewed

 
 
Technology

From Carnegie to Claude: AI Wealth Could Reshape Philanthropy

Anthropic, the maker of the Claude chatbot, will commit 5 percent of its shares to a new public benefit fund. ANDREJ IVANOV / AFP
Anthropic, the maker of the Claude chatbot, will commit 5 percent of its shares to a new public benefit fund. ANDREJ IVANOV / AFP
TT
TT

From Carnegie to Claude: AI Wealth Could Reshape Philanthropy

Anthropic, the maker of the Claude chatbot, will commit 5 percent of its shares to a new public benefit fund. ANDREJ IVANOV / AFP
Anthropic, the maker of the Claude chatbot, will commit 5 percent of its shares to a new public benefit fund. ANDREJ IVANOV / AFP

A wave of American wealth created by artificial intelligence companies could produce some of the largest charitable sums since the era of Andrew Carnegie and John D. Rockefeller, raising questions about where the money goes.

Anthropic, the maker of the Claude chatbot, has said it would commit 5 percent of its shares to a new Anthropic Public Benefit Fund, a stake that could be worth around $100 billion if the company reaches the $2 trillion valuation investors are eyeing for a stock market debut expected as early as November, reported AFP.

Combined with pledges by its seven co-founders to give away 80 percent of their equity and a program matching employees' donations, Anthropic says roughly a fifth of its shares will be dedicated to public-benefit purposes.

"This might be one of the biggest stories in corporate philanthropy ever," said Michael Moody, a professor of philanthropic studies at Indiana University's Lilly Family School of Philanthropy.

The OpenAI Foundation, the nonprofit that controls ChatGPT maker OpenAI, holds a 26 percent stake in the company that was valued at about $130 billion when it was restructured a year ago, more than the endowment of the Gates Foundation.

Nan Ransohoff of payments firm Stripe has described the moment as a "third wave" of American philanthropy, after the industrial fortunes of the late 19th century and the internet billionaires of recent decades.

- 'Dies disgraced' -

Carnegie argued in his 1889 essay "The Gospel of Wealth" that "the man who dies thus rich dies disgraced," and by his death had given away the bulk of his fortune, about $350 million.

Carnegie transferred what remained into a trust, which has continued to make grants for more than a century.

Rockefeller's money built universities and funded medical research and public health campaigns around the world.

Railroad magnate Cornelius Vanderbilt gave comparatively little, apart from the $1 million that founded the university bearing his name.

Anthropic is setting up its fund as a limited liability company that it wholly owns, rather than as a nonprofit foundation, which must pay out a minimum share of its assets each year and file public tax returns.

"The trade-off is that decision-making stays with the company and transparency is voluntary," said Sarah Wolfolds, who studies nonprofit organizations at Cornell University's SC Johnson College of Business.

Anthropic has said the fund's equity will not be used for lobbying or political giving and that it will report on its activities.

- 'Where the money is' -

Many Anthropic founders and staff have close ties to effective altruism, a movement that favors causes such as AI safety and global health because of their measurable impact.

"Nonprofits tend to reorient their programs toward where the money is," Wolfolds said, warning that other causes can become neglected.

Moody said the philanthropic world would be "looking closely at which causes they support," including whether research funding is limited to AI.

Laura McGarry, of fundraising consultancy Graham-Pelton, said the charities best placed to benefit are those that already have relationships with the newly wealthy.

"It's really about values alignment," she said, adding that charities that court the newly rich only after their big payday risk looking transactional.

Effective altruism has shaped some donors' giving but has not "dramatically upset the apple cart," she said.

Anthropic has said the fund could make payments directly to groups affected by AI's impact on jobs, without giving details.

Research on cash transfers is encouraging, Wolfolds said, but "the harder challenge may be identifying who has been affected by AI."

Such payments should complement, not replace, unemployment insurance and retraining programs, she said.

- One child, one share -

Elon Musk, the world's richest person, whose SpaceX owns rival AI developer xAI, criticized Anthropic's approach.

"I recommend giving the shares directly to the public," he wrote on X.

As an alternative, Musk pledged one of his SpaceX shares to every American child next year, a gift worth about $10.5 billion.

US President Donald Trump praised the move as fitting the philosophy behind his "Trump Accounts," government-backed investment accounts for children born between 2025 and 2028. Billionaires Michael and Susan Dell have pledged billions to the program.

Technology

'Tech Debt Coming Due': AI-found Flaw Signals Chaos Ahead

Michael Daniel is president of the Cyber Threat Alliance and was a White House cybersecurity coordinator under former US president Barack Obama. BRENDAN SMIALOWSKI / AFP
Michael Daniel is president of the Cyber Threat Alliance and was a White House cybersecurity coordinator under former US president Barack Obama. BRENDAN SMIALOWSKI / AFP
TT
TT

'Tech Debt Coming Due': AI-found Flaw Signals Chaos Ahead

Michael Daniel is president of the Cyber Threat Alliance and was a White House cybersecurity coordinator under former US president Barack Obama. BRENDAN SMIALOWSKI / AFP
Michael Daniel is president of the Cyber Threat Alliance and was a White House cybersecurity coordinator under former US president Barack Obama. BRENDAN SMIALOWSKI / AFP

Companies and government agencies rushed this week to patch two critical flaws in PHP, the programming language that runs a large share of the world's websites.

The vulnerability to potential hackers was found by using a powerful model from OpenAI in one of the most high-profile demonstration of AI's cyber capabilities.

AFP spoke with Michael Daniel, president of the Cyber Threat Alliance and a former White House cybersecurity coordinator under former US president Barack Obama, about what the episode means as AI tools find software flaws faster than ever.

The interview has been edited for length and clarity.

Q: What does the PHP episode tell us?

A: This is the beginning of a pretty chaotic period, given the abilities of the advanced models. I've been saying for years that the logical conclusion of these AI models being able to find vulnerabilities at speed and at scale is that there's going to be a flood of these vulnerabilities.

Rob Joyce [former cybersecurity director at the US National Security Agency] has referred to it as the great cleansing forest fire of old software. I've talked about our tech debt coming due.

The incentives have been for people to keep software as long as possible, and we don't really factor these risks into the cost of having the software. You're not incurring a direct cost on your books for this old software, but in reality you are, only it's an implicit cost until something like this happens.

Q: What should organizations be doing?

A: Preparing to get your software updated, be prepared to keep it updated, and be prepared to implement patches when they come out.

There's going to be a flood of vulnerabilities, a flood of patches and a flood of mitigations, and not all of them will be created equal.

For a lot of smaller organizations that don't have a lot of internal capacity, what you've got to do is shift your organization to be able to accept the vendor updates as fast as you can.

Q: Is every flaw dangerous?

A: Historically, the exploitation rate for vulnerabilities has been five or six percent, and there's no indication that number is going up. But now when bad guys exploit a vulnerability, they're able to do so much more quickly.

It used to take a month or two for a proof of concept to come out, and then another month or two for it to show up in exploit kits (used by bad actors). That's now measured in hours or days.

Q: What does that mean for large companies?

A: Most businesses essentially ran a zero-risk tolerance for breaking things, so they would test the hell out of a patch before they put it into production. Now they don't have that time.

If the risk of breaking something by putting this patch into place is one percent, but the risk of being attacked could cost me 50 million dollars...now I've got an interesting trade-off.

Q: Are advanced AI tools reliable at finding real flaws?

A: There's a lot of noise. The PHP one is a prime example of a real one, but there's an awful lot out there that are either not real or are on software that's never exposed to the internet. We've got to improve our triage system.

Q: Where does this lead us in two or three years?

A: We're going to turn these AI tools on software that hasn't been deployed yet, so one would hope we start to deploy software that has fewer vulnerabilities to begin with. You burn out all the old stuff, and it makes way for new growth that's more secure from the beginning.

It's not like this solves our cybersecurity problem, but it does change the nature of it going forward.

Technology

CEO: Amazon Cloud Capacity Sold Out Through 2027

FILE PHOTO: Amazon CEO Andy Jassy speaks during an Amazon Devices launch event in New York City, US, February 26, 2025. REUTERS/Brendan McDermid/File Photo
FILE PHOTO: Amazon CEO Andy Jassy speaks during an Amazon Devices launch event in New York City, US, February 26, 2025. REUTERS/Brendan McDermid/File Photo
TT
TT

CEO: Amazon Cloud Capacity Sold Out Through 2027

FILE PHOTO: Amazon CEO Andy Jassy speaks during an Amazon Devices launch event in New York City, US, February 26, 2025. REUTERS/Brendan McDermid/File Photo
FILE PHOTO: Amazon CEO Andy Jassy speaks during an Amazon Devices launch event in New York City, US, February 26, 2025. REUTERS/Brendan McDermid/File Photo

Amazon CEO Andy Jassy said on Friday that Amazon Web Services' cloud computing capacity is largely sold out through 2027, with significant commitments extending into 2028, on growing demand for AI infrastructure.

In a video podcast with Andreessen Horowitz co-founder Ben Horowitz and general partner Erik Torenberg, Jassy said Amazon is struggling to keep pace ⁠with the insatiable ⁠demand from AI labs and enterprises.

Earlier this year, Amazon had said it would spend about $220 billion in capital expenditure for 2026.

According to Reuters, Jassy said the ⁠company "could invest a lot more and it wouldn't be a bad use of capital," considering cloud-computing capacity for 2026 is already sold out.

"2027 is largely sold out. We always leave a certain amount for on demand and our inference service bedrock, or EC2, but largely sold out ⁠all ⁠the big chunks. A good chunk of 2028 is sold out," he said.

EC2 is an AWS service that provides cloud servers for running applications, while Bedrock gives customers access to AI models to build generative AI applications.

Amazon is scheduled to report third quarter results on October 29.