Artificial intelligence security concerns are substantive, but major AI developers currently show limited signs of slowing the pace of model development, according to a BMO note published Friday.
Citing its AI expert, BMO said concerns expressed by industry leaders about AI security reflected the importance of aligning advanced systems with human interests before potential problems become difficult to reverse.
At the same time, BMO questioned whether developers including Anthropic and OpenAI would materially reduce the pace of development, pointing to the continued introduction of increasingly capable models.
BMO Highlights Risks From Open-Access AI Models
According to BMO, recent hacking incidents involving artificial intelligence had generally resulted in limited harm. However, the broker said freely available models could increasingly be exploited by malicious actors.
BMO noted that users have demonstrated methods of bypassing safeguards and prompting AI models to produce outputs that do not comply with their intended restrictions.
AI security has received increased attention following comments from Anthropic chief executive Dario Amodei calling for a coordinated slowdown in the development of advanced models because of potential risks associated with the technology.
According to the supplied material, several AI industry leaders have supported the proposal, while AI security was also discussed during a US-China summit this week.
US and China Could Increase AI Regulation, BMO Says
BMO expects the United States and China eventually to introduce tighter AI regulation, with particular attention given to open-source and open-access models rather than closed systems.
The firm’s AI expert said potential regulations could require approval before open-access models are released more widely to the public.
According to the expert, both governments could have an interest in restricting access to some AI models. BMO’s analysis suggested that such a regulatory environment could increase the pricing power of major closed-source developers, including Anthropic and OpenAI.
These remain expectations rather than confirmed regulatory measures.
BMO Expects AI Token Prices to Continue Declining
BMO expects token prices to continue falling, which it said could support increased use of AI tools.
The firm projects that token prices could reach a low around 2030. It expects closed-source providers could subsequently increase prices as customers become more dependent on their products.
These projections depend on future developments in competition, regulation, technology and user demand.
BMO also said Meta Platforms’ position in artificial intelligence had improved as the company increased utilisation of its AI infrastructure. However, the firm assessed Meta as remaining behind Anthropic and OpenAI in AI capabilities.
Computing Demand Expected to Remain a Focus
BMO maintained a positive view on computing hardware associated with artificial intelligence, citing what it described as increasing scarcity of computing capacity as AI usage expands.
The firm said users were finding that AI capabilities were exceeding some initial expectations, contributing to demand for additional computing resources.
BMO also expects the effects of artificial intelligence on employment to become more significant for white-collar roles by 2028.
The firm projects that effects on blue-collar employment could become more pronounced during the 2030s. BMO linked the expected timing difference to Moravec’s paradox, under which physical tasks requiring dexterity can be more difficult for AI and robotics systems to automate than certain cognitive tasks.
The employment forecasts remain projections and will depend on the development, adoption and deployment of AI and related technologies.