Ontario’s Auditor General Shelley Spence dropped a bombshell on May 12, 2026, with a special report revealing that provincial public servants have been widely using unapproved AI chatbots, while at the same time, the government procured medical AI tools that were never properly tested. The findings paint a troubling picture of disjointed AI governance in one of Canada’s largest provinces, where the gap between official policy and everyday practice has turned into a minefield of privacy breaches and potential patient harm.

The report, which zeroes in on the intersection of artificial intelligence, government operations, and healthcare, underscores how the rush to adopt AI has outpaced the safeguards meant to protect citizens. At its core are two interwoven failures: a “shadow AI” epidemic where staff sidestep formal channels to use generative AI, and a procurement system that greenlit medical AI without mandatory validation. Together, they expose systemic cracks that threaten both the integrity of public service and the health of Ontarians.

The Shadow AI Epidemic: Chatbots Without Boundaries

The auditor’s investigation found that Ontario public servants overwhelmingly turned to unapproved third-party AI chatbots—such as ChatGPT, Claude, and other consumer-grade tools—to handle government work. These platforms were used for drafting internal memos, summarizing policy documents, generating reports, and even analyzing confidential data, all without formal authorization or data protection reviews. In many cases, employees shared sensitive, non-public information with external servers, completely bypassing the province’s IT security policies.

This phenomenon, often called “shadow AI,” is not unique to Ontario. Across the world, workers frustrated by slow internal processes or unaware of restrictions quietly adopt AI tools that promise productivity boosts. But in a public sector context, the risks are magnified. The auditor highlighted instances where chatbots were fed personal health information, proprietary research, and cabinet-level documents—data that could end up stored on foreign servers or used to train future models, with no control over its fate.

One senior privacy analyst familiar with the report described the situation as “death by a thousand clicks.” The analyst, who requested anonymity because they weren’t authorized to speak publicly, explained: “When you have hundreds of public servants copying and pasting sensitive data into a free chatbot, you’ve essentially opened the back door to your entire data ecosystem. It’s not a vulnerability—it’s an active leak.”

The report notes that Ontario’s own policies require departmental approval for any AI usage, along with privacy impact assessments. Yet, enforcement was virtually nonexistent. The auditor’s team discovered that many staffers saw no difference between approved enterprise tools and publicly available ones, and often assumed that because a tool was popular, it was safe. In one cited incident, a public servant used an AI assistant to translate internal documents containing sensitive legal advice into a foreign language, unwittingly sending the text to a cloud service headquartered in a country with no meaningful data protection laws.

Medical AI Tools: Procured but Never Proven

If the chatbot free-for-all raised privacy alarms, the findings around medical AI struck a more visceral nerve. The auditor identified several AI-enabled medical devices and decision-support tools that had been purchased by Ontario’s health agencies and hospitals without undergoing thorough independent validation. These tools—ranging from diagnostic imaging algorithms to patient triage systems—were integrated into clinical workflows on the strength of vendor claims alone, bypassing the rigorous safety checks that are standard for traditional medical technologies.

“We’re talking about algorithms that can influence whether a patient gets a MRI, a prescription, or even a spot on the surgery waitlist,” said Dr. Lena Morrison, a patient safety advocate who reviewed the public summary of the report. “If these models were never tested on Ontario’s diverse population, they could easily misdiagnose, under-diagnose, or exacerbate inequities.”

The report specifically flagged that several procured tools lacked evidence of accuracy across different demographic groups—a critical failure given Ontario’s multicultural makeup. For example, one dermatology imaging tool purchased for use in northern telehealth clinics was validated by its manufacturer exclusively on light skin tones, with no testing on darker skin. Yet it was deployed without local adjustment, risking misdiagnosis of conditions like melanoma in Black and Indigenous patients.

Equally concerning, the auditor found that procurement contracts often classified AI tools as simple “software upgrades” rather than medical interventions, which allowed them to slip through existing health regulatory frameworks. In one case, a predictive analytics platform used to identify high-risk mental health patients was acquired through a basic IT requisition, without clinical review by the Ministry of Health. The platform later generated false positives that, according to internal notes reviewed by the auditor, may have led to unnecessary hospitalizations.

Governance Gaps: A Systemic Breakdown

The twin failures stem from a governance vacuum that the auditor characterized as “urgent and unacceptable.” Ontario lacked a clear inventory of AI tools in use, leaving policymakers blind to the scope of deployment. There was no mandatory training for civil servants on AI risks, no consistent auditing of algorithms, and no mechanism for citizens to know when an automated system made a decision affecting them. The report noted that these gaps violate the spirit—and often the letter—of Ontario’s own Digital and Data Strategy, which calls for responsible and transparent use of emerging technology.

The auditor’s office sampled 15 government ministries and found that only two had even rudimentary AI usage policies, and none actively monitored compliance. This stands in stark contrast to the EU’s AI Act and emerging frameworks in other jurisdictions that demand risk assessments and human oversight. “Ontario is firing first and aiming later,” remarked Mark Thompson, a former CIO of a Canadian Crown corporation, in an interview. “A decade ago we would never dream of deploying unvetted software in critical areas. With AI, we’ve forgotten every lesson we learned about IT governance.”

Particularly damning was the finding that senior management, including some directors and assistant deputy ministers, were among the heaviest users of unapproved chatbots. This trickle-down disregard for rules signaled to rank-and-file employees that policy was optional—a cultural problem that the report says will take much more than a new directive to fix.

The Cost of Convenience

The human and financial costs are already surfacing. The auditor’s report references a 2025 privacy breach at a Toronto-area hospital where a staff member used a generic AI assistant to summarize patient notes. The data, which included identifiable health records, was exposed through the AI provider’s prompt cache, eventually appearing in a public dataset used by researchers. The incident, previously unreported, triggered an investigation under the Personal Health Information Protection Act and resulted in a class-action lawsuit that is still working through the courts.

Financially, untested medical AI carries the risk of misallocation. Ontario spends over $70 billion annually on healthcare, and even a small percentage of that wasted on ineffective or harmful AI could mean hundreds of millions of dollars down the drain. More immediately, the province now faces potential regulatory fines and litigation costs that could have been avoided with upfront due diligence.

Looking Forward: A Tall Order for Reform

The auditor made a raft of recommendations, including:

  • Creating a mandatory province-wide inventory of all AI systems in use, with regular updates.
  • Issuing binding directives that require privacy impact assessments before any AI procurement or deployment.
  • Establishing a dedicated AI oversight office within the Ministry of Government Services, with the authority to audit and sanction non-compliant entities.
  • Requiring all medical AI tools to undergo verification by a new independent health technology assessment body before any public dollars are spent.
  • Launching an urgent public awareness campaign to educate civil servants about the dangers of shadow AI.

The government’s response, appended to the report, accepted most recommendations in principle but offered few concrete timelines. This has drawn criticism from opposition members. “We need more than promises—we need a legislative framework with teeth,” said MPP Jane Park, the NDP’s digital services critic. “If the Premier wants to call Ontario a leader in AI, the first step is to make sure our own house is in order.”

Experts outside government stress that the solution isn’t to ban AI outright. “You can’t put the genie back in the bottle,” said AI governance consultant Amir Hossein. “Instead, you need to give people safe, fast alternatives. Create a library of pre-approved, privacy-compliant AI tools that are easy to access and train everyone on how to use them. As long as the approved path is slower than the shadow path, people will take the shadow path.”

For healthcare, the stakes couldn’t be higher. The report’s release coincides with a broader international push to regulate medical AI. The World Health Organization published updated guidelines in 2025, and the U.S. FDA recently finalized its framework for AI/ML-based medical devices. Ontario’s failure to keep pace threatens not just patient safety but also the province’s ambition to become a global hub for health innovation.

The auditor’s message is blunt: Ontario’s AI free-for-all must end. For Shelley Spence, this report is a repeat of themes she’s tackled before—government overspending, lack of oversight, and the slow churn of bureaucracy. But this time, the consequences are measured not just in wasted dollars, but in eroded trust and, potentially, lost lives.

As Ontarians wait for their government to act, the shadow AI continues to hum beneath the surface. Every unapproved query and unvalidated algorithm is a gamble—one that the auditor general has now laid bare for all to see.