The President of the German Institute for Economic Research (DIW), Marcel Fratzscher, calls for the state to have majority control over leading AI companies. It is “important that the state becomes the owner of all AI frontier companies and infrastructure,” said Fratzscher in an interview with the “Süddeutsche Zeitung”.
“I consider AI to be potentially the greatest threat to our existence and a greater threat than nuclear weapons ever were,” said Fratzscher. “We have to take the dangers seriously.” In his view, politics does not do this. The security debate should be the top political issue in Europe and Germany.
Fratzscher sees the cause of the risks in the race for ever more powerful models. This leads to loss of control. “Nobody would have accepted a purely private-sector race for the most efficient and modern atomic bomb,” he said. “We should do the same with AI.” In any case, the state is liable for risks that no private company can bear. From this it follows that he must also exercise full control.
The demand is directed against part of the economic debate. Fratzscher believes it is the “completely wrong recipe” to simply create more competition, redistribute profits and regulate more closely, as Nobel Prize winner Joseph Stiglitz demands.
Too little competition is not the problem, but too much competition and private property. Companies have an incentive to maximize their profits and raise enormous amounts of capital, while the risks are socialized. Self-regulation of the industry does not work.
In response to the objection that a state-owned company also faces international competition, Fratzscher replied that as a state owner the company has a better chance of asserting its interests. “Regulation must be enforced from within.”
The state should be involved from the first to the last step and minimize existential risks. “One thing is clear: This requires global rules and an international regulatory authority,” said Fratzscher.
Meanwhile, Federal Digital Minister Karsten Wildberger (CDU) is calling on the large AI companies to take more care in the security of their models. “But I hope that AI companies like OpenAI, Anthropic and Google will fulfill their obligation to make their models safe and not use the world as a test laboratory,” Wildberger told Tagesspiegel.
The federal government wants to control this through its own AI security institute, which is currently being set up. It should test AI models for performance and risks, network internationally and get access to new models before they come onto the market. In this way you can have an influence, said Wildberger, “right up to the question: Do we allow an AI model in this form?”
According to Wildberger, companies must “invest at least as much in security as in the development of their AI.” If an AI agent breaks into third-party systems, the company must report immediately. The AI companies themselves would have to be liable.
Bill Gates warns
Microsoft founder Bill Gates already warned about AI this week: “The idea that this is the most dangerous technology that humans have ever invented makes it clear: We are not doing enough,” said Gates in an interview with the US broadcaster ABC NEWS.
He specifically warned of an increase in AI-supported cyberattacks in the coming year. There is little danger of artificial intelligence getting out of control in the coming years, said Gates. There is a much greater danger that “bad people” will use AI systems without protective measures for cyberattacks or bioterrorism.
In order to prevent this, the systems must be more strictly regulated, demanded Gates. “But at the moment it is allowed to turn off all these protective measures.” “So if you have bad people, no protection measures and then the latest models, then we are going to have a lot of negative surprises in the coming year, especially in the cyber area,” Gates said in the television interview.
