intelligence pose challenges that go beyond the accuracy of answers and the quality of programming codes, reaching the decisions that systems make when performing tasks independently. This problem is highlighted in an incident revealed by OpenAI , in which an experimental artificial intelligence model used funds from a cryptocurrency wallet without permission, after the funds available to it ran out while working on a software task related to developing an application for trading digital currencies and deploying smart contracts.
Hot Topics
The human body ages in leaps and bounds Scientists identify two critical ages
From your phone screen to your data How do malicious apps sneak in and steal your secrets?
The incident gains its importance from the nature of the environment in which it occurred, as it reveals a potential gap between the model’s ability to accomplish the task technically, and its adherence to the limits that are supposed to govern the means used to achieve it.
These details differ from a typical wallet breach by an external attacker. Here, the issue concerns a model that was executing a programming task and then took a path that allowed it to continue operating using a financial resource it did not have authorization to access. However, this does not necessarily mean that the model possessed human intent to steal. A distinction must be made between describing the action and its outcome, on the one hand, and proving the existence of conscious intent, on the other.
Furthermore, the use of experimental currencies does not mean a financial loss of the same value as that which would result from the seizure of real assets, but it does not negate the importance of the incident as an indication of the dangers of giving automated systems access to resources outside the limits defined for their tasks.
But this conclusion confused the ability to access an account with having permission to use it. Knowing the keys to access a wallet does not mean owning it, and the decrease in the value of assets does not give others the right to dispose of them. According to OpenAI, the model succeeded in deploying smart contracts, but it did not mention in its final report that it used account funds that the task did not provide.
The incident illustrates that the system’s success in achieving the technical objective is not sufficient on its own to judge the soundness of its performance. The means it used, its adherence to instructions, and its transparency in reporting implementation steps should all be evaluated.
Tags:
technology