Major financial institutions have begun imposing strict restrictions on AI-powered coding agent tools operating on employee devices, a move reflecting growing concerns related to cybersecurity and software supply chains. In this context, press reports stated that Goldman Sachs has restricted the use of tools such as Anthropic's Claude Code and GitHub's coding assistants for developers, with the restrictions limited to bankers and traders, while exempting some financial computer developers known as "Strats".
These moves come amid increasing warnings from tech sector officials, including Amjad Musad, CEO of Replit, who indicated that running intelligent coding agents directly on employee endpoints represents a structural security risk to companies and governments. Musad explained that campaigns to restrict these tools are expanding within the financial sector to mitigate risks associated with supply chain attacks and the exposure of sensitive credentials.
Software Exposure Risks and Credential Protection
Relying on AI agents on user devices leads to issues that already extend beyond the banking sector, including the risk of long-term access keys and software credentials being exposed to reading or reuse by malicious actors. These challenges have pushed some providers toward mandating the use of short-lived tokens and forcing vendors to abandon old keys.
Security alerts issued by tech companies such as Replit point to previous incidents where coding tool extensions distributed via digital stores were exploited to breach systems, confirming that an institution's mere trust in a developer does not prevent dangers if the agent receives malicious instructions or pulls contaminated software dependencies.
What is Changing Practically for Institutions?
The restrictions imposed by major institutions do not mean a complete halt to the use of AI in programming, but rather call for a radical change in operating environments. Modern technical guidelines recommend moving the work of AI agents from local employee devices to dedicated cloud environments controlled by the organization, while separating credential management, audit logs, and approvals.
Experts emphasize that companies need to impose strict control over the scope of permissions and what agents can access, rather than just choosing who is allowed to use those tools, amid expectations that more institutions will adopt similar cloud isolation models to protect their sensitive infrastructure.