Overview
- A joint probe by federal and three provincial privacy commissioners, released Wednesday, found OpenAI broke Canadian privacy law when training the first ChatGPT model.
- Investigators cited overcollection from public websites, a lack of valid consent or clear notice, inaccurate responses about people, and weak ways for users to access, correct or delete their data.
- OpenAI has added filters that mask personal details, limits on training data, tools to stop the chatbot from revealing private information about public figures, and a retention and deletion policy that now applies to current models.
- Regulators set deadlines that require clearer in‑product notices within three months and, within six months, better data exports, challenge paths, protections for retired datasets and tests to block requests for children’s details, with quarterly compliance reports to verify progress.
- Lawsuits and public concern after February’s Tumbler Ridge mass shooting continue to pressure the company as regulators urge Parliament to update privacy laws for AI.