Source: Global News

OpenAI committed to improved protections after a joint federal and provincial investigation found the company engaged in “overly broad” collection of personal information to train its early ChatGPT models. Privacy Commissioner Philippe Dufresne and counterparts in Quebec, British Columbia and Alberta said the probe reinforced an urgent need to modernize Canada’s privacy laws for the AI era. Dufresne told reporters at a press conference in Ottawa that OpenAI launched ChatGPT without fully addressing known privacy issues, exposing Canadians to potential harms such as breaches and discrimination. Investigators identified the crawling or “scraping” of publicly available online information for GPT-3.5 and GPT-4 training as lacking transparency and consent and producing factual inaccuracies, and found Canadians lacked recourse to access, correct, or delete data used in training.

The report criticizes OpenAI for rushing products to market and says that violated accountability requirements under the federal Personal Information Protection and Electronic Documents Act (PIPEDA). OpenAI has agreed to additional safeguards including a personal-data retention policy, significant limits on information used for future training, bilingual transparency measures, and continued compliance monitoring, though the company disagreed with the commissioners’ findings. British Columbia Privacy Commissioner Michael Harvey said ChatGPT may be unable to comply with B.C.’s privacy law as currently written because it forbids reliance on implied consent for third-party data. Alberta’s Diane McLeod and others urged new guardrails while noting some AI development in Canada already adheres to current privacy laws, and federal AI Minister Evan Solomon reiterated that modernizing PIPEDA remains a priority.