Coinbase CEO Brian Armstrong rejects calls for a new AI regulator

Armstrong believes that market incentives encourage AI companies to create safe products without the need for an additional regulatory framework.

Coinbase CEO Brian Armstrong has rejected calls for a new AI self-regulatory body, arguing that existing laws already provide sufficient protection against harmful AI.

His remarks come just a day after the crypto exchange revealed that AI now writes more than 95% of its code, more than double the figure the company announced earlier in the year.

Armstrong says existing laws already cover AI risks

It all started with a proposal from Google DeepMind CEO Demis Hassabis on July 14, in which he called for the creation of a federally overseen standards body to test and certify cutting-edge AI models before deployment.

Artificial general intelligence could arrive within a few years, he says, with increasingly capable models potentially introducing cybersecurity and biological risks as well as a host of national security concerns. He therefore proposed a public-private organization, like the Financial Sector Regulatory Authority, which would initially carry out voluntary reviews before eventually moving to mandatory tests for the most advanced AI systems.

Reaction from Hassabis’s peers was swift, with tech entrepreneur Chamath Palihapitiya calling the framework “pretty well-reasoned, and OpenAI’s Sam Altman describing it as “a thoughtful proposal.” Microsoft CEO Satya Nadella also weighed in, calling it an “important piece” and adding that the goal should be to avoid “any model that breaks the world.”

However, Armstrong disagrees, saying such arrangements often create a dual approval process that requires companies to satisfy both state regulators and industry bodies. He insisted that AI needs neither an SRO nor a government oversight body, since, so far, no harm has been caused that cannot be compensated.

“Why design regulations around a hypothetical problem,” the crypto chief asked. “Existing laws that prevent fraud, award damages when victims suffer harm (tort), UDAP (unfair and deceptive acts and practices) laws, etc. provide broad protections if one of the border labs publishes a model that causes harm.”

Additionally, he pointed out that AI developers also have a strong commercial incentive to release safe products, as users will most likely avoid tools they view as dangerous.

You might also like:

Coinbase is doubling down on AI across its business

Armstrong’s interest in AI is not a passing fad, given that the company he leads has deeply integrated the use of AI into its processes.

This was revealed by a colleague of his, Rob Witoff, Coinbase’s head of platform, who recently told Cointelegraph that between 95% and 100% of the crypto exchange’s code is now written by or with large language models, more than double an estimate shared by the company in February of around 40%.

Recall that in May, the exchange announced a 14% reduction in its workforce, with the intention of reorganizing around smaller, more experienced teams with AI at the center of its operations. According to Witoff, the use of AI on Coinbase varies depending on the task, with sensitive areas such as crypto always subject to detailed human review, while internal prototypes can be built almost entirely through automation.

Other crypto companies, including Gemini, Crypto.com, Kraken, Messari and Dune, have also reduced their workforce this year while expanding the use of AI.

But this rapid adoption of AI hasn’t come without some setbacks, as seen earlier this month when Coinbase was forced to investigate an AI-generated notification that incorrectly reported the result of the FIFA World Cup match between Norway and Brazil before the match had even started.