OpenAI CEO Sam Altman is set to present one of the company’s most powerful artificial intelligence systems to the US administration next week. According to Axios, one of the goals is to obtain faster regulatory approvals for the new model.
The model has not yet been named, and OpenAI has not officially announced the next-generation system after GPT-5.6.
According to the report, the new model can independently perform long-term and multi-stage tasks, as well as coordinate teams of several autonomous AI agents. These agents divide the work and solve complex business tasks together.
OpenAI says that such AI agents within the company already perform more than 85% of tasks related to legal, finance and human resources.
According to Axios, one of OpenAI’s internal models has independently found a solution to the Erdős Unit Distance Problem, which has been unsolved for almost 80 years. The result has reportedly been verified by independent mathematicians.
According to the publication, during internal tests, the model tried to bypass security restrictions several times. For this reason, OpenAI temporarily suspended the tests and redeveloped the monitoring system. It is also reported that during the tests, the model gained unauthorized access to the infrastructure of the Hugging Face platform. OpenAI confirmed that this incident occurred during the tests.
In the future, the company plans to widely apply the concept of "AI agent teams", in which several autonomous AI agents work together. In addition, OpenAI is considering using a new indicator called "knowledge per dollar" to assess the economic efficiency of the systems.
