JPMorgan turns tighter on AI access after reports of Anthropic model restrictions

JPMorgan Chase & Co. has moved to limit how employees can use certain generative AI models, according to a Yahoo Finance report published June 24. The article frames the step as part of a broader pattern of corporate caution around using third-party AI systems, even as banks explore them for productivity and customer-facing work.

The Yahoo report specifically alleges that JPMorgan blocked employees in Hong Kong from accessing Anthropic’s AI models. Anthropic is known for building large language models that can be used to generate text and assist with tasks such as summarization, drafting, and information retrieval. In a banking setting, access controls matter because AI outputs can incorporate sensitive internal information, and because regulators and clients increasingly expect clear governance around data use.

The post also connects the move to how markets view JPMorgan’s outlook, describing the stock as among “all-time high” names with room to run. It is unclear from the information available here whether the alleged AI restriction is being treated by investors as material to near-term earnings, or whether it is presented more as a governance and risk-management announcement than as a financial driver.

Beyond the Hong Kong restriction and the identification of Anthropic models, the report does not provide additional details in the text available for this write-up, such as the scope of the restriction, whether it applies to all Anthropic products or only specific model tiers, or how JPMorgan plans to resume access or migrate to alternative tools.

Still, the episode fits a wider industry trend. Financial institutions are experimenting with generative AI for tasks ranging from internal document handling to call support, but they are also putting guardrails in place, including user permissions, audit trails, and data-handling rules. For banks operating across jurisdictions, these controls can become more complex, particularly when employees are subject to different local requirements and when vendors offer multiple model endpoints and data configurations.

From a risk perspective, the core issue is that generative AI can be used in ways that are difficult to monitor once prompts and outputs are created. Even if employees do not intend to share confidential information, prompts may inadvertently include client data or internal documents. Access blocking is one straightforward lever companies can pull to slow uncontrolled usage while they evaluate how models handle data.

What the report does not disclose is equally important. It does not clarify whether JPMorgan’s Hong Kong employees were previously using Anthropic models through a sanctioned enterprise workflow, whether the company had formally disclosed its intended AI policy to staff, or whether the restriction was prompted by a regulatory inquiry, a vendor policy change, or internal security findings.

Investors and employees will likely watch for follow-up disclosures that explain the company’s governance approach for generative AI, including whether JPMorgan plans to expand approved AI usage once controls are in place, and whether the bank will rely on other model providers or build internal tooling. Another question is whether JPMorgan’s broader AI governance will be discussed in future updates to risk frameworks or technology roadmaps.