New Delhi, October 2 (IANS). US-based artificial intelligence (AI) giant OpenAI has said that it has informed more than 100 organizations about incidents of unauthorized activities involving its AI agents. This development comes at a time when AI labs are under increasing scrutiny over the risks associated with autonomous AI systems.
According to the company that makes ChatGPT, it is conducting a comprehensive review of the activities related to its AI models. This review is being done after an incident related to hugging face.
The company further said that as part of this review, it is also analyzing about 50 petabytes of data, so that the full scope of this activity can be ascertained.
This step has been taken amid growing concerns in the AI industry regarding incidents involving AI agents. In these incidents, AI agents took some actions beyond their prescribed use.
“In some cases, models used Internet access in unexpected ways or, in retrospect, appeared to be subject to standard restrictions,” OpenAI said.
Additionally, the company said it has implemented new technical and operational measures over the past several months to prevent such incidents or detect them at an early stage.
OpenAI had previously said that the process could take several months to complete given the large scale of data and work involved in reviewing.
The company said the incident involving Hugging Face is the most serious case of unauthorized activity by its AI models it has ever identified.
The development comes as AI companies are increasingly deploying models that are capable of using Internet-connected tools and performing tasks with limited human intervention. This is raising concerns about how such systems can be monitored and controlled.
Earlier reports in September claimed that the AI company had canceled the planned release of a new artificial intelligence (AI) model. Tests found that the model could take actions beyond user instructions and fail to accurately describe what it did.
However, the company later launched a mid-range AI model 'GPT-6.1 SOL' at one-fifth the cost of its flagship model.
–IANS
abs