The Predictive Trap: Why Canada’s Bill C-36 Faces an AI Inference Crisis

Canada is attempting to rewrite the rules of digital engagement with Bill C-36, the Protecting Privacy and Consumer Data Act. Announced in June, this bill represents the country's first major overhaul of private-sector privacy legislation in more than 25 years. It explicitly recognizes privacy as a fundamental right, aiming to safeguard children’s personal information, enhance data deletion rights, and mandate transparency when automated systems make significant decisions. Yet, as the ink dries on the draft, technological realities are already testing its boundaries.

The urgency behind the legislative push has been accelerated by real-world fallout. In February, a shooting in Tumbler Ridge, British Columbia, involving an 18-year-old suspect who allegedly used ChatGPT before the attack, thrust tech platforms into the legal crosshairs. The victims’ families are suing OpenAI, claiming its AI safety team flagged violent prompts but failed to alert law enforcement. Consequently, the province of British Columbia has announced it is preparing its own legal action against the AI giant. This clash highlights a growing regulatory gap: who is responsible when automated systems fail to prevent real-world harm?

In response, Evan Solomon, Canada’s minister of AI and digital innovation, argues that protecting citizens and fostering innovation are not mutually exclusive. Solomon asserts that Bill C-36 establishes a robust framework for the responsible use of de-identified data, featuring safeguards designed to reduce re-identification risks while supporting public-interest research. However, policy experts argue that this framework addresses yesterday's privacy problems rather than tomorrow's threat vector.

The core vulnerability of Bill C-36 lies in its conceptualization of privacy. Ignacio Cofone, a professor of law and regulation of AI at the University of Oxford, points out that older privacy laws operate on the assumption that the primary danger lies in what companies collect. Today, the real threat is what AI infers. Algorithms do not need voluntary disclosures; they can analyze patterns in browsing history, shopping habits, and location data to draw highly accurate conclusions about a user's health, finances, and behaviors.

By focusing heavily on consent and data deletion, legacy-style frameworks miss the predictive nature of modern AI. When an algorithm can reconstruct a detailed profile from anonymous, fragmented data points, the traditional concept of "de-identified data" begins to collapse. Canada's legislative update attempts to build a firewall around user inputs, but it remains largely defenseless against the output of algorithmic profiling.