BMA warns of bias, hacking with AI use
The Bermuda Monetary Authority is warning that artificial intelligence tools being used in local finance could increase risks if not properly controlled, including data breaches, discrimination and decisions made with little human input.
In a new discussion paper released last week, the BMA outlined how AI is already being used in Bermuda’s insurance, banking and fintech sectors, to assess creditworthiness, detect fraud and even evaluate job applicants. But it stressed that these tools can create major challenges.
“If AI models are not carefully designed and tested,” the report said, “they may perpetuate or even amplify historical discrimination, unintentionally affecting specific groups of people.” This could happen, for example, if an algorithm denies loans or underwrites insurance based on biased training data.
Cybersecurity is also a concern. AI systems are vulnerable to attacks that can “manipulate or poison models” to give false results, a serious risk when used in fraud detection or claims processing.
The BMA also pointed to “automation bias”, where firms overly trust the output of machines without properly questioning the results. This could lead to unfair decisions if no one understands how the AI came to its conclusion, according to the report.
To address this, the regulator is calling for greater transparency, clear governance and regular audits of AI systems. “Senior management and boards must understand AI’s impact,” the paper said, and firms should be ready to explain how decisions were made.
The BMA is seeking public feedback through September 30 as it prepares to craft new guidance or rules. The goal is to ensure AI is used “responsibly, ethically and in the interest of financial stability”, before problems arise.
Responses should be submitted using the online survey form at bma.bm or directly to policy@bma.bm with the subject line “Discussion Paper – The Responsible Use of Artificial Intelligence in Bermuda's Financial Services Sector”.
• For the full report, see Related Media
You must be Registered or to post comment or to vote.
