This week in AI research: U.S. lawmakers push to ban a Chinese AI disruptor’s app from government devices

The global AI landscape continues to heat up: U.S. lawmakers are zeroing in on Chinese AI disruptor DeepSeek with a possible government-device ban, while open-source challenger Hugging Face fires back at proprietary tools. OpenAI is targeting education and the public sector—both in the U.S. and through a newly forged partnership with SoftBank in Japan—just as Google unveils a powerful new Gemini upgrade. Meanwhile, quantum computing edges closer to practical applications in finance and biomedicine, and Amazon revitalizes Alexa for the generative AI era. Read on for the full rundown.

1. U.S. moves to ban DeepSeek app from government devices

Source: WSJ

Lawmakers intend to introduce a bill on Thursday that would ban the use of DeepSeek’s chatbot application on government-owned devices. This move follows security concerns that the app might share user data with the Chinese government. Drafted by Representatives Darin LaHood (R-Ill.) and Josh Gottheimer (D-N.J.), the bill mirrors the approach taken to restrict TikTok on federal devices and marks the beginning of broader efforts to limit Chinese-owned apps in the U.S.. “This should be a no-brainer in terms of actions we should take immediately to prevent our enemy from getting information from our government,” Gottheimer said, according to the Wall Street Journal.

Last week, Senator Josh Hawley had introduced the Decoupling America’s Artificial Intelligence Capabilities from China Act (S. 321), which sought to restrict U.S. exports, imports, and investments in AI and generative AI technologies with China by banning related technology transfers and R&D collaborations. It threatened significant penalties of up to 20 years in jail or $1 million in fines, or both.

Why It Matters: This latest bill highlights escalating U.S. scrutiny of foreign AI apps on national-security grounds. A potential ban signals that cutting-edge Chinese AI platforms—like DeepSeek—will face tougher hurdles entering or operating in U.S. public institutions.

2. Hugging Face Launches Open-Source Rival to OpenAI’s “Deep Research”

Source: Hugging Face announcement

On February 4, Hugging Face introduced “Open Deep Research,” an open-source framework designed to replicate much of OpenAI’s proprietary “Deep Research” functionality—such as web-based task automation and multi-step reasoning—and make it freely accessible. By making use of an agentic framework that includes web browsing and file-reading capabilities, Open Deep Research achieves a 55.15% score on the GAIA benchmark, compared to Deep Research’s 67.36%. The platform uses a “code agent” approach to express actions, reducing token usage and enabling more complex, multimodal tasks. Developers are invited to contribute additional features—like vision-based browsing—to expand the framework’s capabilities. It is not the only open source option. Jina AI CEO Han Xiao had also launched an AI research agent almost immediately after OpenAI announced Open Research.

Why It Matters: By offering a free alternative to OpenAI’s premium agent-based tools, Hugging Face and other players could help foster an open-source ecosystem that continues to chip away at the market share of AI labs that invest significant resources into launching new features, only to have them copied by open source versions.

1. OpenAI rolls out ChatGPT for education and government

Source: Reuters

OpenAI is launching a version of ChatGPT tailored for the California State University system, benefiting around 500,000 students and faculty. They also introduced “ChatGPT Gov” for U.S. government agencies, offering secure access to its genAI systems. Early concerns about AI-assisted cheating have reportedly eased, according to the article. The move demonstrates OpenAI’s strategic push to integrate large language models into educational and public-sector workflows nationwide. Rival Alphabet has announced a $120 million investment in AI education programs and is introducing its forthcoming Gemini chatbot to students’ Google accounts.

Why It Matters: By offering domain-specific AI tools to academics and government, OpenAI  aims to solidify its foothold in key public institutions. Last week, it announced ChatGPT Gov, designed to provide U.S. government agencies with secure access to OpenAI’s advanced models.

2. OpenAI and SoftBank partner on AI in Japan

Source: AI Business

OpenAI and SoftBank have formed a 50/50 joint venture called SB OpenAI Japan to develop and market AI solutions tailored for Japanese businesses, with SoftBank planning to invest $3 billion annually to deploy OpenAI’s technologies across its group companies. Affiliated firms will receive priority access to models like ChatGPT Enterprise and specialized agent products, and integration is expected in key sectors such as telecommunications, finance, and at Arm (another SoftBank subsidiary) to boost productivity. The venture includes Cristal Intelligence, a secure AI platform for major Japanese corporations, and is linked to the larger Stargate project—an AI infrastructure effort with an initial $100 billion U.S.-based investment (potentially rising to $500 billion over four years) involving Oracle and other stakeholders, rather than solely the U.S. government. While these details capture the essence of the partnership, aspects like Stargate’s funding and Cristal Intelligence’s scope are often oversimplified, underscoring the need for more nuanced reporting.

Why It Matters: The collaboration underscores growing international alliances to scale AI globally. SoftBank’s capital and OpenAI’s technology could point to broader AI adoption in Japan where AI has generally trailed other nations. An article from Nikkei notes that 84% of firms in the U.S. and 85% in China use AI, while only 47% do in Japan, citing J.P. Morgan research.

3. Google unveils Gemini 2.0 Pro with advanced reasoning and huge context

Source: Google

Google has introduced Gemini 2.0 Pro, a model with advanced coding performance, robust reasoning, and a 2 million token context window. Two companion variants—Flash Thinking and Flash-Lite—offer faster or more cost-effective alternatives, all running on Google AI Studio and Vertex AI. According to Google, these upgrades can handle entire codebases and large documents in a single session, indicating the company’s push to maintain its edge in an increasingly competitive AI market.

Why It matters: This matters because major players like Google and OpenAI are vying to lead on features such as coding efficiency and large-scale context handling, and each new release raises the stakes in the race to dominate AI.

4. Interest in quantum builds for drug discovery, finance, climate forecasting and beyond

Source: University of Ontario

Summary: Bank of Montreal (BMO) is joining IBM’s quantum network alongside 50 other financial institutions, aiming to accelerate tasks like climate-risk forecasting and fraud detection. Meanwhile, a University of Toronto and Insilico Medicine study used a hybrid of quantum computing and AI to discover potential drug candidates targeting mutant KRAS—a protein previously considered “undruggable.” Together, these developments signal how quantum computing is moving from theory to application in both finance and biomedicine.

Why It matters:As global banks, technology giants, and academic labs invest heavily in advanced systems, the race to achieve practical quantum advantage is rapidly intensifying.

5. Amazon’s Alexa gets a generative AI overhaul

Source: Reuters

Summary: Amazon is preparing the largest update to its Alexa voice assistant in more than a decade, incorporating generative AI to enable multi-turn conversations and proactive actions on users’ behalf. The revamp, rumored for release after an internal go/no-go decision on February 14, may feature a monthly fee of $5 to $10, although Amazon initially plans a free preview. Company documents point to ambitious goals for Alexa as a daily utility across hundreds of millions of devices, but also acknowledge risks like “hallucinations” — erroneous chatbot answers — that could affect user trust.

Why It Matters: This development matters because Amazon’s move underscores how industry leaders, including Apple and Alphabet, are acknowleding their once groundbreaking conversational AI systems were aging. They now are aiming to embed more advanced AI features into everyday products and services in a post-ChatGPT era.

6. Researchers create open rival to OpenAI’s o1 ‘reasoning’ model for under $50

Source: TechCrunch

Summary: A joint team from Stanford and the University of Washington has unveiled “s1,” an open-source model with reasoning performance comparable to OpenAI’s o1 and DeepSeek’s R1, trained for less than $50 in cloud compute. It is available on GitHub where it currently has 1.7k stars. By distilling “thinking steps” from Google’s publicly accessible Gemini 2.0 Flash Thinking Experimental, researchers fine-tuned the lightweight Qwen base model (from Alibaba’s AI lab) on just 1,000 carefully curated questions. The result: an AI that scores highly on math and coding benchmarks using minimal resources and a simple approach to scaling its “thinking time.” The researchers outlined their methodology in an arXiv preprint.

Why It Matters: This underscores the shifting competitive landscape in AI, where distillation can rapidly clone or approximate expensive, proprietary models at a fraction of the cost. In the coming months and years, distillation will likely lead to increasingly powerful “edge” AI that can run on laptops, smartphones, providing a more distributed genAI reality, which initially was highly centralized.