AI News Report — 2026-07-02 (Thu)

AI news analysis report (last 12 hours)

## 1. [Overview] The US Department of Commerce has lifted export restrictions on Anthropic, and tensions between the government and AI giants have begun to ease. SoftBank resumes negotiations for a $10 billion loan backed by OpenAI stock, and the race for funding continues. Regarding DeepSeek, there have been reports of cases in which AI-generated ransomware has actually been activated, and the risk of AI misuse has come to light.

## 2. [Anthropic trends] The Department of Commerce lifted export restrictions on Anthropic's AI models, and the restart of Fable/Mythos was realized after receiving government approval. The Trump administration lifted regulations after concerns about cybersecurity, which has had a positive impact on Amazon and Broadcom stocks. Palantir's Karp criticized Anthropic and OpenAI's token fee model, saying "something is fundamentally wrong." This is seen as a turning point from confrontation with the government to cooperation.

## 3. [OpenAI trends] Reuters reported that SoftBank has resumed negotiations for a $10 billion loan using OpenAI shares as collateral, with relaxed terms. Barron's points out that OpenAI and Anthropic should rush to IPO, raising questions about the sustainability of their fundraising. Infrastructure development for the government is also progressing, with GPT OSS models becoming possible to run on AWS GovCloud.

## 4. [Google AI trends] Gemini Spark's agent-based assistant is now available for macOS, and its feature is that it can directly handle local Mac files. Google is also testing a function for Gemini AI to recognize the outside of the car for electric cars, and is moving forward with its expansion into the automotive field. A summary of AI-related updates for June was also published on blog.google.

## 5. [Trends of other AI companies] Meta has decided to sell surplus AI computing resources to external parties, similar to SpaceX, and will enter a new business field that will compete with AWS and Google Cloud. At the same time, around 8,000 people were being laid off late at night, and Malay Mail and other media reported that there was widespread opposition and anxiety within the company. xAI releases beta version of Voice Agent Builder to developers. DeepSeek was named in a case where AI-generated browser ransomware exploited the Chromium API, and NVIDIA announced that it has reduced the inference cost of DeepSeek V4 by one-fifth through software improvements.

## 6. [Remarks and trends of AI celebrities] - Sam Altman (OpenAI CEO): Contributed to the Financial Times and discussed ways to make AI safe for everyone. There is also mention of the G7 AI standard formulation proposal. - Dario Amodei (Anthropic CEO): Business Insider reported on the existence of an engineer who supported the White House response amid "Cold War"-like tensions with the government. We are also continuing to make proposals for employment replacement using AI. - Mark Zuckerberg (Meta CEO): Refutes Bill Gates' statement that "only four jobs will remain in the age of AI." Meanwhile, Meta himself has laid off around 8,000 people, and there have been a number of reports pointing out the contradictions between his words and actions. - Elon Musk (xAI Founder / Tesla CEO): Musk denied the WSJ report about SpaceX's AI device prototype (thinner than an iPhone), calling it "completely false." - Yann LeCun (Meta Chief AI Scientist): Criticized xAI as a "failure" and argued for the existence of an AI bubble. There is also mention of confusion within Meta. - Geoffrey Hinton (AI Godfather): Says AI breakthroughs require freedom, not government scripts. - Jensen Huang (NVIDIA CEO): NVIDIA's frugal culture and its ripple effect on AI manufacturing have become a hot topic. Broadcom and NVIDIA's $100 billion AI race is also attracting attention.

## 7. [Breakthroughs and innovations in AI technology] Technically noteworthy is the fact that NVIDIA's software improvements have reduced DeepSeek V4's inference token cost to one-fifth. It was also reported that a team of Harvard dropouts developed a chip exclusively for Transformer, and secured investment and a $1 billion order from TSMC. An explanatory video regarding ARC-AGI-3 has also been released, and interest in general intelligence benchmarks continues.

## 8. [AI Regulation/Policy/Social Impact] Lawfare is analyzing the Department of Commerce's lifting of export restrictions on Anthropic as a move that could shake the foundations of the U.S. AI industry. Courts are still divided on whether copyright use of AI learning data constitutes fair use. Practical questions regarding the use of AI are increasing in the law firm industry, and WSJ also highlighted concerns from the perspective of privacy protection.

## 9. [Integrated analysis: What is happening now] The relationship between the government and AI giants is moving from "confrontation to a transactional truce." The lifting of Anthropic's export restrictions is a symbol of this, with Fortune calling it a "necessary truce." On the other hand, as evidenced by criticism of the token fee model (Palantir) and IPO pressure (Barron's), market doubts about the sustainability of the revenue model remain deep-rooted.

While Meta is transitioning to an AWS-type business model of selling surplus computing resources externally, it is simultaneously undergoing large-scale layoffs, making it the company with the most acute tension between AI investment and employment. In the case of DeepSeek, cases of AI-generated ransomware abuse have been demonstrated, demonstrating once again that the democratization of technology is inextricably linked to security risks.

On the chip infrastructure front, NVIDIA continues to cut inference costs and invest in dedicated Transformer chips, and the race to improve computational efficiency is progressing in parallel with geopolitical export restrictions. Conflicts within the industry are becoming more public, with statements by celebrities highlighting Zuckerberg vs. Gates and LeCun vs. Musk.

## 10. [Future points of interest] 1. Progress in financing Anthropic/OpenAI (SoftBank loan negotiations, IPO observation). 2. Will the external sales business of Meta's surplus computing resources be officially announced, and how will it be explained in terms of its consistency with layoffs? 3. Movement to strengthen security regulations and guardrails in response to AI-generated ransomware cases.

[Overview] The biggest move is OpenAI's negotiations to offer the US government the equivalent of a 5% stake, a move aimed at avoiding regulatory pressure. Anthropic is expanding into the pharmaceutical and government fields with the lifting of export restrictions and the announcement of Claude Science. DeepSeek's low-cost model continues to threaten the strongholds of Anthropic and OpenAI.

[Anthropic trends] The US Department of Commerce has lifted export restrictions on Anthropic's AI models, and Fable 5 has returned to the US government after security concerns were resolved. Announced a tool for pharmaceutical R&D called Claude Science, aiming to advance into the rare disease treatment field. In collaboration with Norton Genie, Claude has also been incorporated into cyber fraud protection. The fact that China's low-cost models are catching up with Anthropic and OpenAI in its home market indicates a deterioration in the competitive environment.

[Trends in OpenAI] Following the FT report, multiple media outlets reported that OpenAI is proposing to provide the Trump administration with a 5% stake (equivalent to approximately $42.6 billion). The aim appears to be to ease political pressure in Washington. Reuters, Bloomberg, and CNBC all followed suit, and the facts are generally confirmed. Sam Altman also calls for a US-led global AI safety forum.

[Google AI trends] Gemini Spark is now available for macOS, and a low-cost, high-throughput image generation model called Nano Banana 2 Lite has also been released. On the other hand, there are moves to restrict Meta's Gemini API access due to compute constraints, and infrastructure constraints are becoming a point of friction in the industry. There are also reports that DeepMind's Demis Hassabis is close to reaching an agreement with the government to formulate AI standards.

[Trends of other AI companies] Meta has launched a strategy to sell surplus compute as a cloud business to compete with AWS, Azure, and Google Cloud. There are also reports of moves to convert AI power plant assets into cash. xAI released ``Voice Agent Builder,'' which allows you to build a voice AI agent with no code, and SpaceX showed a prototype of a consumer AI device, but Elon Musk himself denied reports that the company is an ``iPhone rival.'' With the help of Nvidia, DeepSeek has reduced the cost of V4 tokens by one-fifth and is also offering job openings for developing AI agents with cybersecurity capabilities, while concerns about in-browser ransomware caused by DeepSeek-generated malware have also been reported. Microsoft has made 365 Copilot's Service Agent generally available.

[AI celebrities' statements and trends] Sam Altman (OpenAI CEO) led a 5% equity offer to the US government and advocated for a global AI safety forum. Dario Amodei (Anthropic CEO) warned that AI will surpass humans within a few years, and AWS CEO Matt Garman criticized the idea of replacing junior engineers with AI as "one of the stupidest things I've ever heard." Elon Musk (xAI Founder/Tesla CEO) denies reports about SpaceX's AI devices as "utterly false." Under Jensen Huang (NVIDIA CEO), Nvidia has announced a policy of collecting a portion of the revenue from cloud customers, and is also expanding adoption in the robotics and HPC fields in South Korea. Mark Zuckerberg (Meta CEO) disagrees with concerns about job losses due to AI, saying that AI can lead to job creation even after cutting 8,000 jobs. Palantir CEO Alex Karp criticized Anthropic and OpenAI's token pricing.

[Breakthroughs and innovations in AI technology] Insilico Medicine and Takeda announced a $600 million AI drug discovery partnership, pointing to an expanding $50 billion medical AI market that includes AlphaFold technology. The fact that Nvidia's software update has reduced the inference cost of DeepSeek V4 by one-fifth could be a watershed in the inference cost competition. The United Nations report also mentions AI breakthroughs in the medical field.

[AI regulation/policy/social impact] The US FTC has pointed out that AI bias measures may violate consumer protection laws. Some analysis suggests that the data center boom is creating new antitrust issues. OpenAI's proposed government stock offering itself is becoming a central theme in policy debates over the relationship between AI companies and the state. The new form of AI one-man companies has emerged as an unresolved issue under corporate law.

[Integrated analysis: What is happening now] The central component is the reorganization of the relationship between the US AI giants (OpenAI, Anthropic) and the government. OpenAI's proposal to offer a 5% stake is a move to ease regulatory pressure through political deals, and Anthropic's lifting of export restrictions and the revival of Fable 5 can similarly be read in the context of easing tensions with the government. At the same time, the physical constraints of compute and power have come to the fore as competitive axes, and Meta's sale of surplus compute, Google's access restrictions, and Nvidia's cloud revenue cut policy all reflect this constraint. DeepSeek's cost advantage from China (1/5th the token cost with Nvidia backing) continues to put pressure on U.S. companies to compete on price. In the statements of celebrities, the pessimism of Amodei's camp and the optimism of Zuckerberg, Garman, and others are at odds, and the discourse surrounding the employment impact of AI has not converged even within the industry.

[Future points of interest] - Will OpenAI's proposal to provide government shares reach a formal agreement, and how will competitors such as Anthropic follow suit or compete? - To what extent will DeepSeek's cost competitiveness put pressure on Anthropic/OpenAI's pricing strategy? - How will restrictions on compute access between Meta and Google affect supply chain friction within the industry?