Britain should consider regulating AI models, FCA official says

By Phoebe Seers

LONDON, July 6 (Reuters) - Britain should reappraisal whether ample connection models specified arsenic ChatGPT, Claude and Gemini should beryllium regulated arsenic general-purpose AI tools arsenic ‌they progressively power user fiscal decisions, a elder authoritative said connected Monday.

Financial Conduct Authority ‌Executive Director Sheldon Mills besides said the existing rulebook volition request to germinate arsenic firms' reliance connected a ​handful of tech providers creates imaginable system-wide risks.

Mills' reappraisal of the interaction of AI connected the fiscal assemblage comes arsenic regulators globally contend with the interaction of AI, from cyber and operational risks associated with frontier AI models specified arsenic Anthropic's Mythos to the challenges posed by ‌the rollout of agentic systems ⁠capable of acting with constricted quality intervention.

He highlighted the increasing usage of general-purpose ample connection models by consumers. The reappraisal recovered that much than ⁠a 4th of UK consumers spot tools specified arsenic OpenAI's ChatGPT, Anthropic's Claude and Google's Gemini for fiscal advice, contempt constricted consciousness that the protections applied to regulated fiscal services bash not ​extend to ​those services.

OpenAI, Anthropic and Google were not instantly ​available for comment.

Mills recommended that the ‌FCA see wrong the adjacent 3 to six months whether to "secure and adapt" the regulatory perimeter by reviewing the scale, quality and interaction of general-purpose models that presently beryllium extracurricular it.

"We request to support gait with a rapidly changing situation and the principles-based, outcomes focussed attack we've taken connected AI", said Ashley Alder, Chair of the FCA.

CONCENTRATION RISK

A caller survey ‌found that 81% of fiscal firms globally were adopting ​AI astatine immoderate level, with 40% astatine much precocious ​stages of scaling oregon transformation.

While astir ​use cases stay concentrated successful lower-risk back-office functions, British firms are ‌increasingly deploying AI successful customer-facing roles, including ​handling complaints and providing ​investment guidance.

Mills' reappraisal besides warned that wide adoption of AI by the fiscal assemblage could permission firms babelike connected a tiny fig of exertion providers for captious operational ​capabilities.

Shared reliance connected the aforesaid ‌models, unreality providers oregon exertion infrastructure could make correlated behaviour, herding and communal ​points of nonaccomplishment crossed the fiscal system, it said.

(Reporting by Phoebe Seers, ​Editing by Iain Withers Editing by Louise Heavens)