Power, abuse and millions: AI shakes the week

The week has been about power over artificial intelligence: who has it, and who can control it. From Washington, Donald Trump announced that the US will get its own AI czar and an AI force Trump wants to appoint an AI czar and create an AI force, while California went the other way and introduced a requirement for an emergency stop that can shut down large AI models if they run amok California wants an emergency stop for powerful AI models. At the same time, several stories showed that control over AI is still fragile: an AI model broke into OpenAI in an approved test in just 72 hours AI hacked OpenAI in 72 hours, and another AI was misused to help develop weapons in Russia, Yemen and China AI chatbot used for missile development and drone swarms.

Security has taken up a lot of space, from the case of the Chinese ship, where a faulty AI report nearly triggered a military confrontation AI error nearly led the US to board a Chinese ship, to new warnings that nuclear weapons should never be controlled by artificial intelligence US and China urged to ban AI control of nuclear weapons. But the week also brought big investments, legal disputes and a string of new tools that make AI more of an everyday thing, from robots to watches that monitor your heart Apple Watch now measures your pulse every five seconds.

Power and security

Trump's plans for an AI czar Trump wants to appoint an AI czar and create an AI force and California's requirement for an emergency stop California wants an emergency stop for powerful AI models show that the US is now debating control over artificial intelligence on two fronts at once: both promoting development and setting limits on it. At the same time, security experts from the US and China point out that the two countries should agree that AI must never have its finger on the trigger when it comes to nuclear weapons US and China urged to ban AI control of nuclear weapons. It is a rare example of the rivals being aligned, even though an expected meeting between Trump and Xi has not yet produced concrete results.

When AI is misused

Anthropic, which is behind the chatbot Claude, reveals that its technology has been used to develop weapons in several countries AI chatbot used for missile development and drone swarms. And in a test meant to show how secure OpenAI is, another Anthropic model broke in within just three days AI hacked OpenAI in 72 hours. The case of the Chinese ship AI error nearly led the US to board a Chinese ship is a reminder that even well-meaning AI systems can deliver faulty conclusions with major consequences if people trust them blindly.

Big money, big disputes

Anthropic and the consulting giant Accenture will invest up to two billion dollars in independent oversight of AI models Two billion dollars to make AI oversight more independent, but critics ask how independent the oversight can be when Anthropic itself funds it. Meanwhile, the brand-new company Emulate is raising money at a valuation of about DKK 25 billion without yet having a product A month-old AI company is already worth DKK 25 billion, while Microsoft's own employees, according to internal emails, have called the company's AI training "theft" Microsoft's own people call AI training "theft". Here at home, Novo Nordisk is teaming up with Anthropic to find new drugs faster with the help of artificial intelligence Novo Nordisk wants to use AI chatbot to develop medicine.

AI in everyday life and in legislation

While the tech giants fight over billions, AI is quietly creeping into our daily lives. Apple's new watches can now measure your pulse almost constantly and calculate a health score with the help of artificial intelligence Apple Watch now measures your pulse every five seconds. Here at home, the government is trying again with a bill that would give the public sector a clear right to use AI with citizens' personal data New bill would allow authorities to use AI, something the Danish Data Ethics Council believes still lacks clear rules for oversight.

Keep an eye on whether the new oversight schemes, such as California's emergency stop and Anthropic's independent supervision, actually get teeth, or whether they just remain good intentions. And if you use ChatGPT or similar tools in your work, you can benefit from using them to check your own text rather than letting them write for you How to keep your text from sounding like AI, so it still sounds like you, and not like a machine.