Investing.com -- Citi analysts said Tuesday that calls from AI laboratories to slow frontier AI development are likely driven by efforts to influence regulatory frameworks before a major AI incident occurs, rather than solely by safety concerns.

The investment bank stated that AI labs are attempting to guide policymakers toward capability-based regulation before a significant AI-related event prompts broader, potentially less technically informed government action.

Citi noted that slower AI development could reduce lab revenue growth by delaying enterprise adoption that requires further capability improvements. The bank added that pacing would allow competing open-weight and lagging closed-weight models time to catch up to frontier systems.

The analysts pointed to government reaction to an advanced frontier model release in June as an example of the practical impact. From April 7 to September 1, frontier AI intelligence scores rose from 41 to 53, while leading open-weight model scores advanced from 28 to 45 during the same period.

Citi said that while the latest models have widened the gap between frontier and open systems from 7 points to 9 points over the past two weeks, any slowdown in frontier progress would likely narrow this gap.

The bank stated that the capability gap between model potential and actual user leverage slows enterprise adoption and creates a false sense of security about risks posed by AI systems.

Citi reported that recent advancements in Recursive Self Improvement, the ability of AI systems to autonomously improve themselves, have accelerated beyond expectations. The bank noted this development is "starting to happen across the industry," according to industry sources, occurring earlier than the previously anticipated 2028 timeline.