分类： technology

In a sweeping move that escalates tensions in the global artificial intelligence race, the United States has ordered the full removal of AI technology developed by Chinese firm DeepSeek from nearly all federal civilian government systems, according to a public order issued by the U.S. Department of Homeland Security (DHS).

The ban, which took effect at the end of July but was only disclosed to the public this week, prohibits civilian agencies across the federal government—including the Treasury Department, State Department and Department of Justice—from using any DeepSeek tools or any technology built using DeepSeek’s open-source AI models. Notably, the restriction does not apply to agencies under the Department of Defense, a detail that has not been explained by DHS officials.

The action comes amid growing criticism of Chinese AI development from the Trump administration and leading U.S. AI firms, even as DHS has not publicly outlined specific security vulnerabilities tied to DeepSeek. The White House has ramped up rhetorical pressure on Chinese AI companies throughout 2025, with President Donald Trump repeatedly identifying China as Washington’s primary strategic rival in the AI sector, a technology he has framed as critical to long-term U.S. economic and national security. Earlier this year, White House technology advisor Michael Kratsios wrote in an internal memo that Chinese rivals have been engaged in widespread “malicious exploitation” of intellectual property developed by U.S. AI companies.

Two of the biggest U.S. AI developers, Anthropic and OpenAI—both preparing for initial public offerings in the coming months—have publicly accused DeepSeek and other Chinese AI firms of directly copying proprietary core model architectures developed by U.S. companies. DeepSeek has not yet publicly responded to these allegations, and as of Tuesday, requests for comment have been sent to DHS, DeepSeek, and the Chinese Embassy in Washington D.C., with no official statements released by any of the three parties.

DeepSeek has rapidly emerged as a disruptive force in the global AI ecosystem since its first public launch in early 2025, driven by its open-source business model that makes its core models free for most developers to access and adapt. For startups and independent developers, the cost of building tools on DeepSeek models is just a small fraction of the cost of licensing closed, proprietary models from U.S. providers like OpenAI, Google and Anthropic. The cost advantage extends to DeepSeek’s own development: the company revealed last year that it spent just $6 million to train its original flagship model, compared to the roughly $100 million price tag for OpenAI’s GPT-4 training. That announcement sent shockwaves through global tech markets last year, dragging AI-related stock prices down as investors re-evaluated the cost of competing in large model development.

DeepSeek’s popularity has surged in recent months, even amid growing political scrutiny in the U.S. Its free consumer chatbot ranks among the most downloaded apps in Apple’s U.S. App Store, and its models have already been integrated into customer-facing products from major U.S. tech companies including Snowflake, Vercel and OpenRouter. Just last week, the company released its newest updated model, which has garnered widespread praise from developer communities online for outperforming OpenAI’s most advanced tools on common tasks like software coding.

The ban marks the most significant restriction the U.S. has placed on a major Chinese AI provider to date, and it comes as the Trump administration is weighing additional restrictions on Chinese AI access to U.S. cloud infrastructure and venture capital funding. Industry analysts warn the ban could complicate cross-border AI collaboration and raise costs for thousands of U.S. startups that rely on DeepSeek’s low-cost open-source tools, even as the administration frames the move as critical to protecting U.S. technological advantage.

The global artificial intelligence industry is facing mounting pressure over the unregulated advancement of the technology and its potential long-term risks to humanity, pushing two major tech players to make unprecedented public-focused financial commitments aimed at easing public concerns.

Elon Musk, the billionaire entrepreneur who leads rocket firm SpaceX and AI startup xAI, announced via social media on Monday that starting next year, every child residing in the United States will receive one of his personal SpaceX shares. In his post, Musk framed the move as a response to a similar announcement from AI developer Anthropic, noting “I recommend giving the shares directly to the public. This is what I will be doing” and adding that this giveaway will be the first of many similar initiatives. As of publication, no detailed framework has been released outlining eligibility verification, distribution logistics or the exact value of the shares that will be transferred to recipients.

The announcement came just one day after Anthropic, the company behind the popular large language model Claude, revealed that it would earmark 5% of its total outstanding shares for a newly established public benefit fund dedicated to charitable initiatives aligned with public good. The firm added that its board of directors retains the authority to allocate an additional 2% of company shares to the fund at a later date. In a key clarification to address growing concerns over corporate influence in policy, Anthropic emphasized that no assets held in the public benefit fund will be used for lobbying activities or political donations, whether in the U.S. or in international markets.

The two announcements arrive as both firms prepare for or have recently completed landmark stock market listings. Anthropic’s upcoming initial public offering is widely expected to be one of the largest tech debuts in recent history, with industry analysts projecting a potential valuation exceeding $1 trillion. SpaceX made history in June when it closed its IPO, raising $85.7 billion to claim the title of the largest initial public offering in global market history.

These public pledges come at a moment of intense global scrutiny of the AI sector, centered on growing safety fears, widespread concerns over job displacement, and uncertainty about the technology’s long-term impact on global society. Recent data underscores these risks: a 2024 research report from the International Monetary Fund estimates that nearly 40% of all jobs worldwide will be impacted by AI adoption, with that figure jumping to as much as 60% for advanced economies. Already, thousands of workers across the tech sector globally have been laid off as companies adopt AI tools to boost operational efficiency, leading to growing public backlash against unregulated AI expansion.

Safety concerns have also intensified following high-profile warnings from within the industry itself. Last month, a 27-year-old AI researcher named Jacob Coxon resigned from Anthropic, publishing a viral public post warning that AI development was advancing far too quickly for regulators to mitigate risks. In interviews following his resignation, Coxon stated that many AI insiders are “genuinely frightened” by the pace of progress, adding that “if we don’t slow down at the current rate of progress, there is a strong chance that we could all die in the immediate future.”

Coxon’s former CEO, Anthropic founder Dario Amodei, has previously publicly called for a temporary slowdown in cutting-edge AI development to allow for the implementation of proper safety guardrails, a position that has drawn both support and skepticism from industry observers. Amodei’s call has been backed by two of the biggest names in the AI sector: OpenAI CEO Sam Altman and Elon Musk, both of whom have supported calls for industry-wide deceleration, binding government regulation, and independent third-party monitoring of advanced AI model development.

On the policy front, the U.S. government has taken recent tentative steps to address AI risks. Earlier this month, U.S. President Donald Trump signed a “morally binding” safety agreement with the CEOs of all major AI firms, requiring companies to take responsibility for ensuring the safety of their own AI systems. Notably, President Trump has repeatedly refused to support binding legislative regulation of AI development, despite ongoing bipartisan and public debate over the technology’s risks. Following the signing of the agreement, Trump announced the creation of a new White House AI task force, to be led by Director of National Intelligence Jay Clayton. The task force will be tasked with coordinating U.S. government engagement with AI companies, consumers, and religious organizations on AI-related issues.

Leading artificial intelligence developer Anthropic has delivered a 45% jump in third-quarter revenue, reaching $16.8 billion (£12.7 billion) in the three months ending September, according to sources familiar with the company’s investor briefing obtained by the BBC.

The explosive growth figure was shared Wednesday with potential investors during a presentation at Anthropic’s San Francisco headquarters, as the firm behind the Claude large language model lays the groundwork for an initial public offering (IPO) projected to launch before the end of 2024. A company spokesperson declined to confirm or comment on the unpublished financial data when contacted by reporters.

This strong quarterly result extends a pattern of consecutive growth for Anthropic throughout 2024. Revenue climbed from $4.7 billion in the first quarter to $11.6 billion in the second quarter, and the firm notched a second consecutive profitable quarter, according to the briefing materials.

Anthropic’s financial performance, paired with results from industry rival OpenAI, serves as a key benchmark for measuring the broader AI sector’s actual growth trajectory and its capacity to fund the massive computing infrastructure required to advance next-generation model development.

The upbeat earnings update comes at a moment of heightened market volatility across AI-focused and tech hardware stocks, as investor anxiety grows over the enormous capital outlays required to build out AI capacity. On Wednesday, major U.S. technology shares pulled back, with leading semiconductor manufacturers including AMD, Nvidia and Intel recording sharp share price drops.

Market unease spread to Asian exchanges Thursday morning. Shares in Japanese technology investment conglomerate SoftBank fell 9%, while South Korean semiconductor leaders Samsung dropped 8.4% and SK Hynix declined more than 6%. South Korea’s tech-heavy Kospi index tumbled 7.6% in early morning trading.

The recent market declines follow critical remarks from Hock Tan, chief executive officer of U.S. tech manufacturing firm Broadcom, delivered at an industry conference in Singapore earlier this week. Tan argued that many investors hold “extreme expectations” for AI returns that are disconnected from actual market fundamentals, comparing current AI investment sentiment to gambling in a casino where many participants “have no clue” about the underlying risks.

Beyond market concerns over infrastructure spending, the AI industry continues to grapple with widespread debate over model safety. Last month, OpenAI CEO Sam Altman confirmed his firm would not pursue an IPO in 2024, prioritizing work to address outstanding safety risks before moving forward with a public listing. The ChatGPT developer is also preparing for a blockbuster IPO that is projected to value the company at more than $1 trillion.

Anthropic has itself been a prominent voice in calls for more deliberate AI development, with CEO Dario Amodei repeatedly urging the industry to slow the pace of advanced model rollouts to mitigate unaddressed safety risks. Despite these public concerns over sector risks, the company has continued its global expansion, opening a new regional office in Singapore on the same day it held its investor briefing in San Francisco.

A Saudi national and former OpenAI engineer has been arrested and charged by federal authorities in the United States over allegations he planted a hidden listening device in a sensitive internal conference room at the leading AI firm’s offices. Thirty-five-year-old Abdulrahman Alfozan, a legal permanent U.S. resident who had worked at OpenAI for roughly a year, made his first federal court appearance this week in California following his October 9 arrest by FBI agents.

Court documents filed by prosecutors lay out a month-long alleged pattern of planning stretching from mid-August to mid-September this year. According to the official complaint, Alfozan first used a productivity AI tool to gather critical information about the company’s meeting infrastructure, repeatedly querying the system for details about meeting room schedules, common meeting locations for external partner discussions, and daily bookings for the target conference room.

Prosecutors further allege that in September, Alfozan used a chatbot linked to his spouse’s account to ask for guidance on constructing, installing, and remotely accessing a covert recording device. The targeted meeting room, investigators note, was routinely used to host discussions covering highly sensitive material: including confidential technical development updates, commercial strategy, OpenAI’s government partnerships, and private business relationships. Alfozan held no authorization to participate in or access meetings held in this space, the complaint confirms.

Closed-circuit security footage captured Alfozan near the restricted conference room on eight separate occasions throughout September, according to investigators. On September 30, OpenAI staff made the first discovery: a hidden audio recording device, wrapped in toilet paper to conceal it, tucked behind a mounted television in the room.

To identify who placed the device, investigators positioned a surveillance camera inside the conference room. On October 5, Alfozan was recorded entering the room and checking the spot where the listening device had been installed, court documents state. That same day, security personnel found an unattended bag Alfozan left in a separate meeting room; inside, they discovered a second matching recording device, along with the engineer’s wallet and official identification.

In a public statement posted to the platform X, OpenAI confirmed it had leveraged its own proprietary AI models to help analyze the device and trace its origins. The company also stressed that, to date, investigators have found no evidence that Alfozan successfully accessed or removed sensitive core company assets, including AI model weights, proprietary source code, or private customer data. Alfozan’s employment contract with OpenAI has since been terminated following his arrest.

A native of Saudi Arabia, Alfozan was educated in the U.S. and has held legal permanent resident status in the country since 2010, according to court records. His defense attorney William Pernik pushed back on the allegations in a statement to media, noting that Alfozan built his life in the U.S. after arriving years ago as a young person chasing opportunity. Pernik described the charges as entirely out of character for his client, saying the legal team will conduct a full, exhaustive review of all evidence presented by the government to clarify what actually occurred.

Pernik emphasized that under U.S. legal principles, Alfozan is presumed innocent until proven guilty, and the public allegations released so far do not constitute a finding of guilt. He asked the public to refrain from passing judgment until all facts come to light in court.

A detention hearing is scheduled for October 19 to decide whether Alfozan will remain in federal custody while the criminal case proceeds. The FBI’s investigation into the alleged plot remains ongoing. If convicted on the single charge of illegally attempting to intercept oral communications, Alfozan faces a maximum sentence of five years in federal prison and a fine of $250,000.

In a high-profile case that spotlights growing security risks within the global artificial intelligence industry, federal prosecutors from the U.S. Department of Justice have formally charged a Saudi national employed as an engineer at leading AI developer OpenAI with orchestrating a covert surveillance operation inside the company’s San Francisco office.

The defendant, identified as Abdulrahman Alfozan, has been accused of one federal count of using or attempting to use an electronic surveillance device to intercept private oral communications. Court documents outline that Alfozan had joined OpenAI as an engineer only months prior, in July 2025.

OpenAI, the organization behind the world-dominating chatbot ChatGPT, first flagged suspicious activity after internal teams uncovered a well-concealed, sophisticated recording device tucked inside one of the company’s executive conference rooms. According to official complaint details, the device was wrapped in toilet paper and plugged into an unused USB port mounted on the rear of a conference room television — a spot chosen specifically to avoid easy detection. The device was discovered by OpenAI staff on September 30, two full days after prosecutors allege Alfozan installed it, and investigators confirmed the device was encrypted to protect any data it collected.

Remarkably, OpenAI confirmed that it leveraged its own proprietary AI models to trace the origins of the rogue device, ultimately pinpointing Alfozan as the person who planted it. The company immediately notified the Federal Bureau of Investigation (FBI) to launch a formal criminal investigation, and moved quickly to terminate Alfozan’s employment. To date, OpenAI has confirmed there is no evidence that sensitive corporate assets including core model source code, model weights, or private customer data were improperly accessed or exfiltrated during the incident.

FBI Director Kash Patel released an official statement confirming the allegations, noting that Alfozan “allegedly concealed a covert recording device inside his employer’s executive conference room to secretly capture sensitive communications.” Patel added that the stated goal of the operation was to obtain confidential internal information about OpenAI’s cutting-edge AI product development.

Investigative records further outline that Alfozan used common workplace productivity tools and an external chatbot to plan his operation in advance. Prosecutors allege that chat prompts linked to Alfozan include research queries on building, customizing, installing, and hiding audio recording devices, adding remote access functionality, improving live audio capture quality, and encrypting data stored on the device’s memory card.

Prosecutors say that on October 5, roughly a week after the device was installed, Alfozan returned to the executive conference room to check on the covert recorder. Closed-circuit security camera footage from the office confirms his presence at the site, directly tying him to the location of the device. On that same day, Alfozan reportedly left an unattended bag in another section of the OpenAI office building. When the company’s security team inspected the bag, they found Alfozan’s wallet and government identification, along with a second recording device matching the design of the one found in the conference room.

Alfozan made his first federal court appearance on Tuesday, and as of the latest update, it remains unclear whether he has secured legal representation to fight the charges. If he is ultimately convicted on the single federal charge, Alfozan faces a maximum penalty of five years in federal prison and a $250,000 fine.

The incident has sparked new conversations about insider threat risks for high-stakes AI development firms, which hold closely guarded intellectual property that is widely considered among the most valuable commercial assets in the global technology sector.