Experts in artificial intelligence say it’s risky for the Canada Revenue Agency to turn to artificial intelligence to help Canadians with their tax problems when human call-centre employees have shown to struggle providing accurate information.

The Canada Revenue Agency (CRA) is under scrutiny after a newly released report from the Karen Hogan, Auditor General of Canada, revealed deep shortcomings in its call-centre services and cast serious doubt on its plan to rely on artificial-intelligence tools. According to the report, live agents provided accurate information in fewer than one in five (20 per cent) personal-income-tax queries.
The report highlighted that a rule-based chatbot, known as “Charlie,” which the CRA has deployed to handle routine queries, only delivered correct responses about one third (≈ 33 per cent) of the time. Singapore-style generative-AI versions are already being piloted, yet experts warn the underlying human-agent data remains unreliable.
Unlike fully generative systems, Charlie responds from predetermined scripts and cannot synthesise new content, which limits its adaptability to nuanced tax-questions. The CRA began offering a generative-AI chatbot beta on March 24, 2025, available around the clock to handle questions about personal income tax, charitable status and account access.
“It’s essential that the human-agent foundation is reliable before layering on AI,” said Anatoliy Gruzd, Canada Research Chair in Privacy Preserving Digital Technologies. “If your human agents are getting answers wrong, the machine will simply replicate or amplify those errors.”
Professor Adegboyega Ojo of Carleton University echoed the warning, saying the CRA should adopt a hybrid model: chatbots handling straightforward queries, and human experts stepping in for complex or exceptional cases. “AI can ease intense load, but must not replace trained human judgement,” he said.
Experts additionally pointed to government guidelines on AI use. A federal document on generative-AI advises agencies to evaluate risks, monitor accuracy, ensure human oversight and adopt transparency.
Service-delivery advocates say the CRA’s struggles are longstanding. A 2017 audit found that call-centre agents gave incorrect answers almost 30 per cent of the time.
The current report confirms that despite billions in funding and technology investment, performance remains weak and taxpayer trust is eroding. Complex tax-questions, language-barrier issues and high agent turnover all contribute to breakdowns.
In its 2025-26 departmental plan, the CRA announced it would expand its use of machine-learning and AI to detect non-compliance and improve operations.
But critics argue that reducing contact-centre staffing or relying overly on chatbots before agents are adequately trained will worsen service outcomes.
For Canadians seeking help during tax season, the message is sobering: technology alone will not fix service gaps. Until the CRA ensures its human agents are trained, confident and accurate, rolling out sophisticated AI-tools may simply scale up the mistakes. Maintaining public trust will depend on more than fast web-chats—it will depend on reliable support, transparent systems and accountability for when things go wrong.