Google's I/O 2026 developer conference in Mountain View, Calif. on Tuesday.Manuel Orbegozo/Reuters
Some shareholder groups are increasingly concerned about the liability risks that come with the unfettered development of artificial intelligence and are pushing companies to adopt more stringent oversight measures.
Vancity Investment Management in Vancouver is requesting that Google-parent Alphabet Inc. GOOG-T do a better job of preventing AI chatbots from spreading misinformation, while other investors including the Pension Plan of the United Church of Canada want Shopify Inc. SHOP-T to put in place a policy governing responsible AI use.
The investors filed shareholder proposals with Alphabet and Shopify ahead of each company’s annual meeting in June. Both companies have advised shareholders to vote against the proposals.
Alphabet and Shopify declined to comment.
We need to rethink Canadian standards for AI
The risks related to generative AI include biased results and false information, as well autonomous agents that can take unintended but harmful actions, such as deleting computer code. AI has become a focus for some socially responsible investment groups in recent years, including the human rights risks associated with facial recognition technology.
Five of the 10 proposals filed by Alphabet shareholders concern AI, including advocating for more disclosure around the climate and water impacts of building data centres. Meta Platforms Inc. is also facing a handful of AI-related proposals. One requests that the company conduct an audit to assess the risks related to the data used for training AI models, which includes copyrighted material.
Along with Open Mic, a non-profit that works with investors to push for changes at tech companies, Vancity is requesting that Alphabet’s board hire independent evaluators to recommend additional steps the company can take to prevent its AI applications, such as Gemini and AI Overviews, from spreading false information. Google’s AI Overviews, which are AI-generated answers for traditional search queries, reach more than 2.5 billion people each month, while the Gemini chatbot app has some 900 million monthly users, according to the company. Alphabet announced plans this month to further inject AI into the online search experience.
Why AI success depends on culture, not code
Alphabet’s testing has shown that its Gemini 3 model, released last fall, scored 72.1 per cent on a benchmark that measures factual accuracy. While the company hailed the performance as state-of-the-art, it translates into an error rate of roughly 28 per cent. Google has said that pulling from search results to answer questions, such as AI Overviews does, can improve accuracy. A study last fall by AI startup Oumi found that AI Overviews were accurate about 91 per cent of the time.
“The general public just uses the AI Overview, and accepts that as the correct answer,” said Edmond Ho, associate portfolio manager at Vancity. Alphabet has done a decent job at developing internal guidelines and checks for AI development, he said, but external input is missing. He also argued that the company should be more forthcoming with data around the accuracy of its models, beyond brief disclaimers stating that AI can make mistakes.
Vancity said in its proposal that factual errors and hallucinations – misleading, or entirely fabricated information that sounds plausible – from AI exposes Alphabet to legal risks (the company is already subject to lawsuits related to AI-generated fabrications), which could result in financial penalties. “There’s real financial risk for us as investors,” Mr. Ho said.
Alphabet said in filings in response to Vancity’s request that its “multi-layered governance framework” and technical disclosures show that it is already making efforts to clamp down on AI-generated misinformation. The company also said it has used independent auditors to stress-test its models and publishes “essential information” about the findings.
Alphabet is facing another proposal from the Shareholder Association of Research and Education, a Canadian non-profit. Share, as the organization is known, is requesting Alphabet formalize oversight of the human rights risks associated with AI with the board’s audit committee. The proposal was co-filed with investors PFA Pension and Parnassus Investments.
The proposal was motivated by Alphabet’s decision last year to remove its commitment to not use AI for weapons and surveillance applications, and a change to board committee charters dropping oversight of civil and human rights. While this responsibility is spread across the board, according to Alphabet, Share argues this approach leaves a transparency and accountability gap.
“We do see that Alphabet is lagging behind some peers,” said Juana Lee, associate director of corporate engagement at Share, pointing out that companies such as Microsoft and Cisco have assigned AI risk oversight to at least one committee. “We believe it is an integral part of protecting long-term shareholder value,” she said.
Alphabet is recommending that shareholders vote against the proposal, saying that its board is stacked with experts who have “unparalleled technical and scientific depth” and that its board structure allows committees to be flexible when assessing AI-related risks.
It’s not just frontier AI developers such as Alphabet that are facing shareholder proposals, but companies that are adopting the technology as well. Share, on behalf of the United Church of Canada’s pension plan, wants Shopify to implement a policy for responsible AI use to assure investors the company is properly managing risks.
Shopify is a rapid adopter of AI, adding tools for merchants and working with Google and OpenAI to allow consumers to buy products directly through AI conversations. Among other risks, Share noted that AI systems have the potential to make erroneous automated transactions.
Shopify said in corporate filings that the proposal is a “solution in search of a problem.” The company has a code of conduct for employees and tests AI products before launch, among other measures. “We understand AI deeply and build responsibly – grounded in real-world use, not a policy document detached from our practices,” Shopify said in its management information circular.
Proxy advisory firm Institutional Shareholder Services also recommended investors vote against the proposal in a note issued this week. ISS said it is “overly prescriptive and unnecessary at this time” and that Shopify’s broader risk management approach appears to be sufficient.
Ms. Lee said that AI policies are becoming more common. eBay Inc.’s policy, for example, states that its AI systems should not cause unintended consequences or perpetuate bias, and that it will work to reduce hallucinations.