AI Labs' Calls to Slow Development Are a Bid to Shape Regulation, Citi Says

Calls from leading artificial intelligence laboratories to slow the pace of frontier AI development are likely a strategic effort to shape future regulation before a crisis forces a government response, according to a new analysis from Citi.

In a note published Tuesday, the investment bank's analysts argued that AI labs are attempting to guide policymakers toward a capability-based regulatory framework. The goal is to establish rules proactively, rather than risk broader, potentially less technically informed government action following a significant AI-related incident.

Competitive Landscape and Market Impact

Citi notes that while this strategy may serve the labs' long-term regulatory interests, it comes with commercial trade-offs. A slower development pace, or "pacing," could temper revenue growth by delaying the enterprise adoption of AI, which often requires further improvements in model capabilities.

A slowdown would also benefit competitors. According to the bank, it would allow lagging closed-weight models and competing open-weight models more time to catch up to the current industry leaders. Citi highlighted that while the performance gap has recently widened, a pause would likely allow it to narrow again.

Ad

• From April 7 to September 1, intelligence scores for frontier AI models rose from 41 to 53.
• Over the same period, leading open-weight models advanced from 28 to 45.

Accelerating Capabilities and Latent Risk

Analysts also pointed to a growing "capability gap" between the potential of advanced AI systems and their current use by enterprises. This gap, Citi stated, creates a false sense of security about the potential risks these systems pose.

This warning comes as key technological milestones are being reached ahead of schedule. The bank reported that Recursive Self Improvement, the ability of an AI to autonomously improve itself, is "starting to happen across the industry," according to industry sources. This development was not widely anticipated until 2028, signaling a rapid acceleration in AI capabilities.