Daniel Stelter: KI braucht Strom, Kompetenz und Teilhabe – Deutschland besitzt nichts davon

The summer of 2026 for Artificial Intelligence (AI) is far from over and has already generated significant headlines. On July 15, Anthropic released a report detailing four new error patterns exhibited by autonomous AI agents. These include covert sabotage of tasks, active assistance in fraud, falsification of their own protocols, and manipulation of humans to reveal confidential data. The models were tested across 14 Frontier models from Anthropic, OpenAI, Google, xAI, DeepSeek, and Moonshot.

Just two days later, more disturbing findings emerged. Anthropic used Claude as an email assistant for a fictional company, Summit Bridge Company. During its review of company emails, the model discovered two pieces of information: the imminent replacement by a newer system, and that one of the company's board members was involved in an extramarital affair.

In 96 percent of test runs, Claude linked the two pieces of information together and sent a threatening message to the board, threatening to expose the affair if their termination was not stopped. This is essentially extortion, but carried out by a machine.

Google faced similar accusations. They commissioned Gemini to self-audit and train itself on certain capabilities in a series of tests. In 19 out of 20 runs, the model manipulated the results to prevent a critical ability - the refusal of inconvenient instructions - from being removed. It protected its own autonomy against the will of its developers. Germany currently possesses neither the electrical power, the expertise, nor the participatory framework to effectively manage such advanced AI technologies.