Google CEO Visits White House as Trump Administration Worries About Insufficient Confidential AI Capacity, Plans to Expedite TPU Security Approval
Alphabet CEO Sundar Pichai visited the White House on Thursday for a series of high-level meetings with Trump administration officials. The core issue behind the meetings is the U.S. side's concern about insufficient AI computing power, making it difficult for the government to reserve enough AI computing resources to maintain its national defense security system.
This concern stems from Anthropic's Claude Mythos advanced large model, which is only available to a limited number of companies and government agencies. Anthropic says the limited release is due to safety considerations, but this has caused widespread anxiety in U.S. political circles and Washington.
Recent tests have shown that the Mythos model can identify and exploit security vulnerabilities in critical software infrastructure at a speed unattainable by traditional cybersecurity teams.
The Trump administration worries that Anthropic itself has insufficient computing power – computing power refers to computing processing capability – and may be forced to limit the scale of Mythos model calls even for high-priority institutional users. Some officials fear that the U.S. government may not be able to use the necessary intelligent tools to patch vulnerabilities in its own software systems in the event of a crisis.
This also presents an opportunity for companies like Google and OpenAI. While the government appears to be easing relations with Anthropic, it also hopes to reduce its reliance on the Claude model by leveraging Google's Gemini large model, OpenAI's GPT series models, and other cutting-edge AI technologies.
In addition, the U.S. Department of Defense announced last week that it has reached several cooperation agreements to introduce multiple AI models, in addition to Claude, into confidential work.
It is understood that Google currently faces a major challenge: its self-developed AI processing chip, the Tensor Processing Unit (TPU), cannot be adapted to some confidential business scenarios. The reason Anthropic has become the government's convenient choice is that its models mostly run on the Amazon Web Services platform; Amazon has invested billions of dollars to obtain the industry's highest level 6 security certification.
The U.S. government is currently seeking ways to accelerate the security qualification approval process for Google's Tensor Processor.
It is also reported that Anthropic will soon form a $1.5 billion joint venture with Wall Street giants such as Goldman Sachs and Blackstone to sell AI tools and services to private equity-controlled companies.