OpenAI Sounds Alarm on China's DeepSeek AI Threat

OpenAI is urging the U.S. government to consider restrictions on AI models from DeepSeek, a Chinese lab it describes as "state-controlled" and "state-subsidized."

In a submission to the Trump administration’s "AI Action Plan," OpenAI warns that DeepSeek’s AI, including its R1 reasoning model, poses serious privacy and security threats. It argues that Chinese law could force DeepSeek to surrender user data, risking sensitive information and intellectual property theft.

It's unclear exactly which DeepSeek models OpenAI views as problematic—whether it means the open models widely used by companies like Microsoft and Amazon, or DeepSeek’s API. OpenAI previously accused DeepSeek of improperly extracting knowledge from its own models.

The latest claims escalate OpenAI’s criticism significantly by suggesting DeepSeek operates directly under Chinese government influence. Yet, there's no proven link between DeepSeek and China's authorities. DeepSeek originated from High-Flyer, a private quantitative hedge fund.

However, China’s government is clearly paying attention. Recently, DeepSeek’s founder, Liang Wenfeng, met personally with Chinese President Xi Jinping, highlighting growing interest from the state.

Following the initial controversy, OpenAI spokesperson Liz Bourgeois clarified the company's stance. She said OpenAI isn't pushing to restrict individual use of DeepSeek’s models. Instead, they're proposing modifications to U.S. export rules, allowing countries access to advanced American computing resources—provided their data centers avoid relying on potentially risky Chinese technology.

This policy shift aims to safely expand global AI capabilities. The goal, according to OpenAI, is simple: more secure computing power available to more countries, without inadvertently empowering China's surveillance capabilities.

Google Fires Back in AI Wars with Powerful, Pocket-Sized Models

Google has unveiled Gemma 3, its latest open AI models designed to deliver powerful performance on lower-cost hardware, stepping up competition against rivals like DeepSeek.

Gemma 3 offers lightweight AI models ranging from just 1 billion parameters to a sleek 27 billion, optimized for performance on single devices, including phones.

Google claims these models are its most advanced, portable, and responsibly designed open models yet, and says they outperform DeepSeek-V3, Meta’s Llama, and other comparable options.

The standout feature of Gemma 3 is its expanded “context window,” now handling up to 128,000 tokens, significantly boosting its ability to manage larger data inputs.

Alongside this, Google launched ShieldGemma 2, an AI-powered image safety checker built with Gemma 3, aimed at detecting sensitive content like violence or explicit images.

Additionally, Google re-entered robotics, launching Gemini Robotics, a model derived from Gemini 2.0 that interprets natural language commands and adapts to changes in its environment, reviving the company’s interest after scaling back robotics ambitions in recent years.

Safety remains a key priority. Given the risks with open models, Google emphasizes extensive testing, highlighting Gemma 3's rigorous evaluation to mitigate misuse, especially due to its enhanced ability to support larger data inputs of up to 128,000 tokens, catching up with industry leaders.

These moves mark Google's latest push to maintain a leadership role in the rapidly evolving—and increasingly competitive—AI industry.

Microsoft Levels Up Copilot with Immersive 3D Gaming

Here's your punchy, concise summary written using the inverted pyramid style:

Microsoft is diving deeper into gaming, hiring a senior engineer to bring 3D experiences to its Copilot AI platform.

A recent job listing reveals Microsoft seeks a Beijing-based specialist skilled in 3D rendering engines like those used in popular video games.

This hire would expand Copilot, Microsoft's AI-powered chatbot, into immersive gaming interactions.

Microsoft previously hinted at its gaming ambitions with Muse, an AI trained on Ninja Theory’s multiplayer title, Bleeding Edge.

Muse is designed to understand 3D environments and enable short, interactive games within Copilot itself.

Microsoft showcased Muse's capabilities earlier this year, promising interactive, short gaming experiences soon.

Additionally, Copilot has already dipped its toes into gaming, demonstrated by guiding players in Minecraft-style interactions—like crafting swords by checking inventories and recommending missing resources.

Beyond practical advice, Microsoft has also explored "character-based" Copilot personalities to boost user engagement and make AI interactions feel more lifelike.

This gaming initiative highlights Microsoft's broader effort to transform Copilot from a productivity assistant into a fully interactive, entertainment-oriented platform.

For gamers, this could mean more engaging, responsive, and personalized experiences directly inside Microsoft's AI ecosystem.

Google’s Gemini Chatbot May Soon Turn Your Text into Videos

Here's your summarized article, following your guidelines:

Google is secretly preparing new AI-powered video-generation tools for Gemini, its popular chatbot app, according to hidden code found by Android Authority.

The latest Google app update references “robin,” an internal codename signaling upcoming Gemini AI capabilities.

While details remain scarce, experts speculate Gemini might soon integrate Veo 2, Google DeepMind’s sophisticated AI model capable of turning text descriptions into realistic videos.

Veo 2 currently represents cutting-edge tech in AI-generated visuals. Right now, high-quality AI video generation requires significant computing resources and typically remains behind paywalls or limited-access programs.

Adding Veo 2 to Gemini could revolutionize how users create content, making video generation as easy as typing a sentence.

Alternatively, Google might introduce a streamlined version of Veo 2—simpler and less resource-intensive—to better suit Gemini's broader consumer audience.

Another potential (but less exciting) route could involve integrating Gemini with Google’s existing AI tools like Google Vids. Google Vids uses AI for tasks like writing scripts and generating voiceovers, but it's currently geared toward professional users, making it less likely for Gemini’s mainstream debut.

Google hasn’t yet confirmed pricing or availability. Still, AI-driven video generation is typically resource-heavy, indicating such a feature could become part of Gemini's premium subscription service.

Launching this kind of capability would position Gemini as one of the first mainstream platforms to offer high-quality AI-generated video clips directly from users' text inputs, setting a significant benchmark.

Whether exciting or unsettling, Google's move to embed advanced AI video generation into everyday apps like Gemini underscores the tech industry's rapid march toward making once-elite AI tools accessible to everyone.