Google has removed a key commitment from its AI principles that previously pledged not to use artificial intelligence for weaponization, Bloomberg reports.

The company’s AI Principles originally included a clause stating that it would not develop AI applications or technologies “for harm, including weapons.” That wording is now gone.

When asked for comment, Google referred to a blog post.

“We believe that democratic nations should lead AI development, guided by fundamental values such as freedom, equality, and respect for human rights. Companies, governments, and organizations that share these values should work together to create AI that protects people, promotes global growth, and supports national security,” stated Google Senior Vice President James Manyika and DeepMind CEO Demis Hassabis.

Former Google AI ethics lead Margaret Mitchell pointed out that removing the phrase about “harm” could indicate that the company is working on technology capable of killing people.

The change aligns with broader shifts in policy among major AI companies. Previously, Meta opened its AI models to U.S. government agencies and defense contractors, while Anthropic made a similar move.

In January, OpenAI signed an agreement with U.S. National Laboratories to apply AI models in nuclear weapons research.

In August 2024, 200 employees from Google’s AI division, DeepMind, protested the company’s involvement in military projects and called for the termination of defense contracts.

AI Investments

Meanwhile, Google CEO Sundar Pichai announced plans to invest around $75 billion in capital expenditures to expand artificial intelligence.

According to Pichai, these funds will accelerate AI innovation and strengthen Google’s core business sectors.

Capital expenditures refer to funds spent on acquiring long-term physical or essential assets for business operations. It remains unclear how much of the investment is specifically allocated to AI.

In January 2025, Meta CEO Mark Zuckerberg vowed to maintain high AI-related spending despite concerns over the DeepSeek AI model.