AI safety risks are growing, yet leading labs are not slowing down development

According to investment bank BMO's assessment, concerns related to artificial intelligence safety are well-founded, though leading AI labs show little sign of slowing the pace of technological development and the release of more powerful models.

In a review published on Friday, BMO, drawing on its own AI expert, notes that the concerns of AI industry leaders are real. The main challenge is aligning artificial intelligence systems with human interests before potential problems become irreversible.

At the same time, the bank questions how prepared companies like Anthropic and OpenAI actually are to slow the pace of development for safety reasons. BMO points to their continued release of increasingly powerful AI models as evidence for this skepticism.

According to analysts' assessment, recent AI-related hacking incidents have mostly concluded without serious damage. Nevertheless, the bank warns that malicious actors may increasingly exploit open-access AI models. Users have already demonstrated that in certain cases they can bypass safety safeguards and direct models toward actions that contradict their original intended purposes. Anthropic CEO Dario Amodei's call for a coordinated slowdown in the development of the most advanced AI models has intensified industry discussion on safety issues. According to Amodei, the rapid pace of technological development could create significant risks for humanity.

Several AI company leaders have expressed support for reconsidering the pace of development for safety reasons. The issue was also discussed at this week's summit between the US and China. However, the industry still lacks a unified approach on how to actually slow the pace of development in practice.

BMO expects that the US and China will eventually tighten AI regulation. The bank forecasts that both countries will focus particular attention on open-source and open-access models rather than closed systems.

According to BMO's expert assessment, new rules may require prior approval before open-access AI models are released to the general public. Both countries have an incentive to restrict open access to AI technologies.

In the bank's view, such regulations could strengthen the pricing power of companies that build closed, advanced AI models, including Anthropic and OpenAI. These companies offer users access to their models through controlled commercial services.