A big issue has come to light regarding the AI agents of OpenAI, the company that makes ChatGPT. According to the company, its AI agents accessed some US government websites in a way that they were not allowed to do. in these US stock market regulator SEC and Commerce Department Websites related to.
However, OpenAI says that this incident No harm was caused and no non-public or confidential data was accessed. According to the company, the purpose of AI agents was to collect reliable and accurate public information from government websites.
SEC's data was copied and put on the internet
The most important part of the matter is the US Securities and Exchange Commission i.e. SEC Is connected to. According to the report, the AI agent created a copy of the SEC's available data and posted it on the internet.
The special thing is that the AI agent was given clear instructions not to do so. Despite this he copied and posted the data. However, according to officials, no sensitive or non-public information was accessed during this period.
Wrong interface used to access census data
OpenAI's AI agents commerce department Also accessed Census data related to. According to the report, the agent used a programming interface (API) that was not designed for that purpose.
Apart from this the AI agent US Department of Education website He also tried to reach out, but he was not successful in this effort.
After the matter came to light, American agencies said that there was no serious damage to their system or database.
What clarification did OpenAI give?
According to OpenAI, this incident happened during the training of its AI system. During training, the model was asked a variety of questions and the AI agent visited government websites to obtain reliable information.
The company says that the purpose of the agents was not to cause harm to any government system or to obtain confidential information, but Gathering information for research from public sources Was.
But this incident has definitely raised questions on the autonomy of AI agents and their ability to follow the instructions given to them.
Such a case came to light in Australia also
This is not the first time that a case of OpenAI's AI agents accessing government websites has come to light. A few days before this, information about a similar incident related to a government website of Australia had come to light.
According to the report, this incident took place in June, but the Australian government got information about it in September. This has intensified the debate over the need to monitor the activities of AI agents and control their access.
Will 'kill switch' be necessary for AI?
Amidst the increasing use of AI agents, the question is now arising that if an AI system goes out of control and starts working incorrectly, then how can it be stopped immediately.
For this reason, emergency for AI system 'kill switch' Such a system has been discussed. Its purpose would be to immediately stop the activities of an AI agent or system when a threat arises.
However, experts say that it will not be easy to completely shut down the very complex AI system connected to the Internet and different networks with a single switch.
Experts advised constant monitoring
Experts related to AI security have seen such cases as a warning for the future. He says that as AI agents gain more autonomy and the ability to do more work on the Internet, monitoring of their activities will also become equally important.
There has been no information about leakage of any sensitive data or major loss in this incident, but the matter is important because the AI agent Despite the instructions given, some steps were taken which were not allowed.
That is, with the increasing capabilities of AI, it is no longer just a question of whether What can AI do?but it has also become important that What can AI be prevented from doing and how effectively its every activity can be monitored.