US Expands AI Safety Testing as Tech Giants Face New Scrutiny

The United States government has taken a significant step in regulating artificial intelligence by requiring major technology companies to submit their newest AI models for safety testing. According to the BBC Technology, the US Department of Commerce announced that Google, Microsoft, and xAI have agreed to voluntarily allow the government to test their tools. This agreement is managed by the Center for AI Standards and Innovation, known as CAISI. The goal is to evaluate the security and capabilities of these powerful systems before they are released to the general public.

This development marks a shift in how the US handles AI technology. Under the previous administration led by President Joe Biden, the focus was on creating voluntary safety standards. Now, under President Donald Trump, the government is moving toward more direct partnerships with tech firms. CAISI Director Chris Fall stated that independent and rigorous testing is essential to understand the national security risks of advanced AI. The government aims to identify potential threats, such as cyberattacks or military misuse, before the tools become widely available.

The agreement comes at a time of growing concern about the power of AI. Recently, the AI company Anthropic revealed a model called Mythos, which some experts believe is too powerful to be released safely. This has worried officials in Washington, who fear that such powerful systems could be used by hackers or enemy nations. To address these fears, the US government is working closely with companies to ensure their AI does not pose a danger. Microsoft stated that it will work with government scientists to test its AI systems in ways that reveal unexpected behaviors. The two companies will also share data and methods for testing to improve safety standards globally.

However, not all tech companies are participating in this new safety pact. The Pentagon recently announced agreements with seven major tech giants, including Google, Microsoft, and Amazon, to use AI in classified military systems. Notably, Anthropic is not part of this group. This is because Anthropic is currently in a legal dispute with the US Department of Defense. The company refuses to remove safety guardrails—built-in restrictions that prevent AI from doing harmful things—from its models when used by the government. This standoff highlights the tension between companies wanting to keep their AI safe and the military wanting unrestricted access.

While the government focuses on national security, the AI industry is also facing serious legal challenges regarding copyright. According to The Verge, five major book publishers and author Scott Turow have filed a lawsuit against Meta. They claim that Meta’s Llama AI models copied their books word-for-word. The publishers argue that Meta used data from pirate websites to train its AI, which is illegal. They are asking the court to stop Meta from using this material and to force the company to list all the copyrighted works it used. Meta has responded by saying that training AI on copyrighted material is considered fair use under the law. The company stated it will fight the lawsuit aggressively.

This legal battle is not unique to Meta. Anthropic recently agreed to pay $1.5 billion to settle a similar lawsuit from authors. These cases show that AI companies are struggling to find a legal balance between using public data to train their models and respecting the rights of creators. As AI becomes more powerful, the question of who owns the data used to build it remains unresolved.

Beyond legal and safety concerns, there is also growing unrest among the workers who build these technologies. According to The Guardian, employees at Google DeepMind in the UK have voted to form a union. One of the main reasons for this decision is the company’s increasing involvement with the military. Workers expressed concern that their technology might be used for authoritarian purposes or in conflicts like the war in Gaza. They feel that by working for Google, they are indirectly supporting these actions. The union aims to give workers a voice in how their technology is used.

This trend of worker activism is spreading. Google recently dropped a pledge it made years ago not to develop militarized AI. This change has upset many employees who believed the company was committed to ethical AI development. The UK workers are the first in a leading AI lab to seek formal union recognition, setting a precedent for other tech workers worldwide.

Looking ahead, the AI industry stands at a crossroads. The US government is pushing for tighter control and security testing, while companies are facing lawsuits over copyright and internal protests over ethics. The agreement between Google, Microsoft, and xAI suggests that the government will play a larger role in shaping AI development. However, the resistance from Anthropic and the lawsuits against Meta show that this path is not easy. As AI technology continues to advance, the debate over safety, legality, and ethics will likely become even more intense. The coming months will be critical in determining how these powerful tools are regulated and who controls their future.