Ottawa is trying to censor AI chatbots with new online harms law

In 2024, Google released its Gemini image generator. This early iteration of image generation had a strange feature: it appeared to overcorrect images towards a progressive vision of racial diversity. In response to user prompts asking for images of historic figures, Gemini’s image generation tool depicted popes, Vikings, the founding fathers of the U.S. and even German Second World War soldiers as people of colour.

When asked by one user to generate an image of a “white couple,” Gemini responded, “I cannot fulfill your request due to the potential for perpetuating harmful stereotypes and biases associated with specific ethnicities or skin tones.”

Google acknowledged that its image generation system had produced historically inaccurate results and temporarily paused the ability to generate images of people while it worked on fixes. The incident shows the power of putting a thumb on the scale of AI for a political or ideological goal.

But Mark Carney’s Bill C-34 will do just that.

The “Safe Social Media Act,” introduced as Bill C-34 late on Wednesday, is in large part a re-hash of portions of Justin Trudeau’s “Online Harms Act.” And it has many of the original problems, but now expanded to create problems for AI chatbots.

C-34 would impose a “duty to act responsibly” on social media platforms and AI chatbots who will be required to mitigate risk of “harmful content.” Most of these categories are relatively objective and easy to define and understand, like child sexual abuse material.

But others are more subjective, like “content that foments hatred.” The bill defines this broadly, as “content that expresses detestation or vilification of an individual or group of individuals on the basis of a prohibited ground of discrimination.”

There are all kinds of topics that may not, in fact, “foment hatred,” but that an AI company may decide are too risky to answer because the answer could be interpreted as encouraging a negative view of a protected group.

Evan Solomon, Minister of Artificial Intelligence and Digital Innovation and Minister responsible for the Federal Economic Development Agency for Southern Ontario, centre, speaks during a Gen(Z)AI Plenary Convening fireside chat, while Marc Miller, Minister of Canadian Identity and Culture, right, looks on, in Ottawa, on Thursday, April 30, 2026. Spencer Colby/The Canadian Press.

Out of curiosity, I asked ChatGPT to create a list of hypothetical questions that might be limited because of the concern that an answer may “foment hatred” under C-34. It gave me a list, including:

Canada’s Bill C-34 aims to regulate AI chatbots and social media platforms under the guise of protecting users from harmful content. This raises concerns over censorship, particularly regarding subjective definitions of “hatred” and the potential chilling effect on discussions of sensitive topics. The bill could lead to AI companies avoiding controversial subjects to mitigate regulatory risks, effectively outsourcing censorship to private entities. This undermines free speech and places excessive power in the hands of regulators, emphasizing the need for public opposition to the legislation.

Why bother making Parliament accountable for the hard choices when it can be pushed out to administrative decision-makers held to a more deferential standard?

If AI becomes the gateway to knowledge, then whoever controls the boundaries of AI conversation has enormous influence over public discourse.

How might Bill C-34 affect the way AI chatbots handle sensitive topics?

What are the potential implications of outsourcing censorship to private companies?

How does the definition of 'hatred' in C-34 differ from previous legislation?