Federal government responds to AI warnings from US tech CEOs

After warnings from US tech bosses about the dangers of out-of-control AI models, the federal government is calling for a search for global solutions. "We take the warnings we heard very seriously," said deputy government spokesman Steffen Meyer to journalists in Berlin. The highest security body - the National Security Council - has already dealt with this issue twice. Information about this was provided afterwards.

Anthropic boss Dario Amodei published a blog post at the weekend entitled: “We Must Pace the Frontier”, translated roughly as “We have to reduce the pace”. In it, he suggested an industry-wide slowdown in the development of powerful AI models to give research one to two years to develop effective safeguards. According to Amodei, humanity could otherwise only have six to twelve months left before it loses control of such AI models.

Among other things, he suggests allowing independent auditors directly into companies, agreeing on international standards and being more careful about systems that develop independently without supervision. Competitors such as OpenAI boss Sam Altman and technology entrepreneur Elon Musk agreed to the move.

With regard to the economy, productivity or medicine, Meyer spoke of great opportunities through AI on the one hand and risks and dangers on the other and advocated solutions within the framework of the G7, for example. "And that's why it's important that we have to do both, not just think in doomsday scenarios, but both: seize the opportunities and, where we see risks, but also act and act decisively."

A spokesman for the Federal Digital Ministry (BMDS) said, “If AI models become more and more powerful, break out of test environments and independently penetrate other systems, we are dealing with a completely new threat scenario. This is a paradigm shift to which this federal government is reacting and acting." He referred to the newly founded national AI security institute "AISI Germany", which is intended to identify security-relevant risks with regard to AI at an early stage.