Nobel Prize-level AI is very likely to come in 2026! Anthropic put forward six feasible recommendations to the White House, calling on the United States to accelerate AI innovation and risk management at the policy level. At the same time, AI is moving from the lab to the battlefield, and the Pentagon has decided to add agents to operational planning.

By 2026-2027, the level of AI intelligence will reach the level of Nobel Prize winners.

Anthropic’s latest long article once again declares that humanity is close to superintelligence.

Report address: https://assets.anthropic.com/m/4e20a4ab6512e217/original/Anthropic-Response-to-OSTP-RFI-March-2025-Final-Submission-v3.pdf

In the article “Machines of Loving Grace“, Dario Amodei, CEO of the United Arab Emirates, participated in the prediction that by the end of 2026-2027, humans will witness the birth of a powerful AI system.

They have the following features:

Interdisciplinary intelligence: in biology, computer science, mathematics, engineering, etc., at or beyond the level of Nobel laureates.
Use digital interfaces: From generating text, audio, and video, to autonomously operating a mouse, keyboard, and browsing the internet, these systems will seamlessly take over the current digitization of humans.
Long-term reasoning solves complex tasks: AI will act like a high-powered employee, independently completing complex tasks that span hours, days, or even weeks, and proactively seek feedback when needed.
Ability to interact with the physical world: Through digital connectivity, AI is able to control lab equipment, robotic systems, and manufacturing tools.

These predictions are not utopian.

Anthropic’s latest Claude 3.7 Sonnet and Claude Code models have demonstrated significant AI enhancement capabilities and greater autonomy.

In addition, the cutting-edge results of other top experiments in the world are equally exciting.

The Pentagon has also stepped in to accelerate the deployment of AI.

Today, the U.S. military signed an agreement with Scale AI to include AI agents in military planning and combat operations in the future.

Even Microsoft is involved in providing support for large models.

01

U.S. AI Action Plan: Six Recommendations

In the face of this revolution, Anthropic firmly believes that the United States must act decisively to maintain global technological leadership.

To that end, they put forward six recommendations to balance the economic and security implications of AI while benefiting all Americans.

Here are all 6 suggestions:

1. National security testing

Government agencies should establish robust assessment systems to comprehensively analyze the potential impact of AI models on national security at home and abroad. This includes establishing a framework for evaluating standards, building a security testing infrastructure, and forming an expert group to analyze the vulnerabilities of deployed systems.

2. Strengthen export controls

Anthropic advocates for tighter restrictions on semiconductor exports to ensure that the United States and its allies can seize the opportunities presented by powerful AI systems and prevent adversaries from gaining access to the infrastructure that supports powerful AI.

3. Improve lab safety

As AI systems become a key strategic asset, Anthropic proposes to establish a confidential communication channel between AI labs and intelligence agencies, expedite the security review process for industry professionals, and develop a next-generation AI infrastructure security standard.

4. Expand energy infrastructure

To secure its position at the forefront of AI development, Anthropic proposes to set an ambitious target of adding 50 gigawatts of dedicated power by 2027, while streamlining the approval process.

5. Promote the use of AI in government

Anthropic advocates for a comprehensive inventory of government workflows that could benefit from AI enhancements, and tasked agency leaders to drive projects that deliver significant public benefits.

6. Embrace economic change

To ensure that the benefits of AI are widely shared across society, Anthropic recommends modernizing economic data collection mechanisms, such as updating the Census Bureau’s survey methodology, and preparing for the large-scale changes that could occur in the economy.

These recommendations are based on Anthropic’s long-term policy research.

This time, the full recommendations submitted provide further implementation strategies that hopefully provide practical support for the U.S. government to address this technological transition.

02

AI agents are about to join the battlefield

The ambitions and actions of the United States in the field of AI are gradually emerging.

Here, the U.S. military has signed an agreement with Scale AI to apply AI agents to military planning and combat operations.

According to Scale AI’s statement, the project will be led by the company, together with Anduril and Microsoft, to create the U.S. Department of Defense’s “first attempt to fully integrate AI agents in military workflows”.

According to Bryce Goodman, head of the Thunderforge program at DI, “using AI to transform the military decision-making process is critical to maintaining U.S. military superiority.”

“The current military planning process still relies on decades-old technologies and methods, which creates a fundamental gap between the speed of modern warfare and our ability to respond,” he said. Thunderforge brings AI-driven analytics and automation capabilities to operational and strategic planning, enabling decision-makers to work at the speed needed to respond to new types of conflict.”

The goal of the Thunderforge project is also very clear, to help military commanders make judgments faster and more accurately by accelerating information processing and decision-making evaluation through AI agents.

Not only can these AIs support mission planning and campaign development, but they can also assist in theater-level resource allocation and strategic evaluation.

The text makes it clear that while the decision-making process is guided by software, all final decisions are still made by humans.

In particular, the DIU mentioned that these AI agents, as “advanced robots”, can perform sand table deductions, simulate various outcomes for commanders, plan different scenarios, and “refine the proposed action plan”.

The first tests will take place in a real-world environment, with future deployments planned at the U.S. Indo-Pacific Command in Hawaii and the current U.S. European Command in Germany.

The Pentagon plans to roll out the system to all 11 of its joint operations commands after initial deployment.

In the Thunderforge project, the division of labor is very clear.

Anduril will be responsible for providing its Lattice software platform, while cloud computing giant Microsoft will be responsible for processing the actual large language models.

Scale AI’s role is to apply its expertise in the application of AI agent technology and generative AI evaluation.

Alexandr Wang, founder and CEO of Scale, is confident that “our AI solutions will transform existing military operations processes and modernize U.S. defense.”

Resources:

Author：桃子好困
Source：Anthropic预测26年AI智力堪比诺奖得主！美国AI行动计划发布在即，五角大楼紧急布局
The copyright belongs to the author. For commercial reprints, please contact the author for authorization. For non-commercial reprints, please indicate the source.