In this investigation, we saw that, when OpenAI first developed its models, it didn't comply with the legislation. It obtained too much information from Canadians—about all of us—through everything on the Internet, social media and discussion forums. All this was used to train the models. They could obtain confidential information, medical information or information about young people.
There was also misinformation and disinformation. The statements made on social media weren't always true, but the responses could give the impression that they were. We found that transparency in this area was necessary and that tools were needed to determine where a response was coming from.
We also found that the consent process wasn't good enough for users and for Canadians in general. Work was needed in this area. This investigation gave us the opportunity to bring all these factors to OpenAI's attention, and we persuaded it to make these changes.
That said, the current tool is quite different from the one originally launched. This is largely in response to our recommendations.