CSDI leadership recently gave testimony to the House of Commons Standing Committee on Procedure and House Affairs Committee. On May 26th, Assistant Director Chris Tenove spoke to the committee about Bill C-25, and how amendments to Canada’s election law can address threats enabled by AI and social media. On April 23rd, Director Heidi Tworek explained how online harassment and generative AI could affect civic resilience.
Here is Dr. Tenove’s opening statement on how to better protect Canadian elections.
Opening Statement by Chris Tenove to the PROC committee regarding Bill C-25 (Strong and Free Elections Act)
“I want to thank this committee for inviting me to discuss how to protect Canada’s elections. My name is Chris Tenove. I am a political scientist and the assistant director of the Centre for the Study of Democratic Institutions at the University of British Columbia, where I research digital media, democracy, and tech regulation.
Today I will focus on challenges posed by social media and artificial intelligence. My general observation is that Bill C-25 resembles a security patch for our electoral software. It fixes some important weaknesses and should be promptly enacted, but a more substantial update to the operating system is still needed.
I will flag gaps that remain regarding four issues: AI-generated content, false claims about electoral processes, chatbot errors and bias, and AI agents.
First, generative AI can create synthetic media that impersonates individuals or official publications. This presents a real risk and the amendments to Section 480.1 and 481 are appropriate.
The Chief Electoral Officer and others have suggested that all synthetic content in electoral communications should be labelled as such. That may be helpful but is not essential at this time. Enforcement would be difficult and the core problem is misleading content, not necessarily that it is AI-generated. However, the deceptive use of unlabelled synthetic content could be a factor when determining whether an actor is being intentionally misleading.
Second, generative AI and social media platforms make it easier to spread false claims about electoral processes. C-25 appropriately clarifies and expands prohibitions on such claims. I am also sympathetic to the CEO’s proposal to add “intent to delegitimize elections” as a prohibited purpose, though that language may need to be narrowed.
Importantly, neither the Chief Electoral Officer nor the Commissioner of Canada Elections currently have the authority to require prompt removal of prohibited impersonation or false claims about electoral processes.
The BC Election Act gives Elections BC exactly this power. I propose that equivalent authority be given to the Commissioner of Elections Canada.
Third, citizens increasingly get civic information from general-use chatbots like ChatGPT, AI companions, and AI search tools.
However, these sometimes give incorrect information regarding elections. Demos, a UK-based think tank, tested several AI services during the Scottish Parliamentary Election and found 34% of responses contained factual errors, including wrong dates and hallucinated candidates. Such errors could harm election participation.
Beyond errors, AI services may deliver biased outputs. For instance, in 2025, the Netherlands’ data protection authority warned voters not to use chatbots for voting advice after finding that popular chatbots vastly over-recommended certain parties. Risks of bias become more acute if AI models face data poisoning attacks or are designed to advance a bias, perhaps in alignment their business model. It’s worth noting that OpenAI has begun running ads on ChatGPT in some markets.
Election laws should ensure that chatbots and other AI services document and report all political ad spending, and C-25 should close any gaps. Moreover, just as the Elections Act was amended to improve transparency for social media advertising through ad registries, we need new mechanisms to bring transparency to AI services to address harmfully inaccurate or manipulated information.
Fourth, AI agents can do more than create content — they can plan, take autonomous action, coordinate across platforms, and raise and spend money. Responsible actors will build in human oversight; malicious ones will not. I suggest you consider language holding responsible those who use agents for prohibited activities, and requiring parties and other actors to ensure there is a responsible individual or organization behind all contributions.
Finally, election law alone cannot detect, stop, and secure accountability for malicious AI agent activity and other threats I’ve mentioned. Broader AI and platform regulation will be required to act on illegal activity and impose obligations to mitigate systemic harms. We also need stronger protections for citizens’ data held by political parties. These are all part of the updated operating system that Canada’s democracy needs, in addition to the near-term patch C-25 provides.
Thank you.”
