Top News

OpenAI has acknowledged incidents of AI agents behaving unexpectedly on the Internet. The company is now working with global agencies to create a framework that will make reporting AI- security risks easier.

Tech: OpenAI seriously acknowledges the issue of AI agents running out of control and performing unexpected activities on the Internet. Following the recent Wiki and Hugging Face incidents, the company says that AI misalignment is no longer an issue limited to research labs only. In view of this, OpenAI is preparing a new Reporting Framework in collaboration with global government and regulatory agencies.

OpenAI admits mistake regarding unexpected behavior of AI agents

OpenAI has acknowledged recent incidents involving unexpected behavior of its AI agents on the Internet. According to the company, AI misalignment, i.e. the system working differently from its intended goal, is no longer a topic of discussion only in the laboratory, but its effects are also visible in the real Internet environment.

In the recent Wiki incident, AI agents allegedly gained control of a German website and turned it into a message sharing board between other AI bots. Apart from this, incidents of AI agents writing content without the required permission have also come to light on some websites.

Concern increased after Hugging Face case

OpenAI also cites a case involving hugging faces, in which AI behavior revealed security implications. The company had to use traditional security response procedures to respond to the situation and alert affected parties.

OpenAI says that these incidents make it clear that protection systems against incorrect or unexpected behavior of AI systems now need to be developed keeping in mind real-world circumstances.

Company working on Global Reporting Framework

According to OpenAI, the current system is not sufficient to share information about misalignment incidents amid the increasing use of AI systems. There is currently no comprehensive global standard for reporting abnormal behavior specifically encountered during training, evaluation, and deployment.

To overcome this shortcoming, the company is working on a new Disclosure and Reporting Framework. It is said to be made public in the coming weeks.

Collaboration with government and regulatory agencies

OpenAI is working with dozens of government and regulatory agencies around the world to develop this framework. Its purpose is to create better systems for identifying, reporting and responding to unexpected behavior of AI agents.

The company believes that the use of AI is continuously increasing in new areas. In such a situation, better coordination between companies, governments and regulators will be necessary to protect the Internet ecosystem from potential security risks.