Goldman Sachs Halts Employee Vibe Coding Over Security and Productivity Risks
Goldman Sachs has suspended employee use of AI text-prompt software development tools, citing productivity concerns and security risks following an internal trial period.
Goldman Sachs has reportedly directed its employees to halt the use of artificial intelligence tools for building software through text prompts, a practice commonly known as "vibe coding". Over the past few months, the global investment banking giant had permitted personnel to utilize Anthropic's Claude Code and Microsoft's GitHub to independently create custom tools for tracking trading activities, modeling market shifts, and generating content.
According to reports citing internal developments, the decision follows comprehensive reviews that raised concerns regarding productivity limitations and potential operational risks. The sudden restrictions have sparked frustration among various staff members, particularly after Marco Argenti, the firm's chief information officer, publicly praised non-engineer employees such as traders and bankers in July for leveraging AI tools to autonomously develop workflow enhancements.
In a statement to the press, Goldman Sachs characterized the experimental phase as an evaluation period aimed at assessing productivity factors and risk management frameworks. Representatives noted that data science departments remain exempt from these new limitations, allowing specialized technical teams to continue utilizing advanced development platforms.
Industry analysts point out that security vulnerabilities frequently accompany software generated entirely through automated prompts. A recent study by Veracode revealed that code produced by artificial intelligence tools successfully passed security compliance testing in only 56% of instances. Despite scaling back internal deployment, Goldman Sachs maintains a strong financial connection to the sector, serving as a prospective underwriter for Anthropic's upcoming public offering.
