Why are companies restricting AI access to employees

Companies are being prompted to rethink how freely their workforce can use artificial intelligence (AI) tools to improve productivity as concerns grow around data security, confidentiality and the protection of proprietary information.

Goldman Sachs is the latest to tighten access. The Wall Street bank has restricted most employees, including bankers and traders, from using Anthropic’s Claude Code and Microsoft’s GitHub coding tools, limiting access to engineers, Bloomberg reported, citing an October internal guidance.

For months, the bank allowed employees across functions to experiment with the tools. Traders used them to monitor client trading flows, model market changes and positions, and create written and graphic content, the Bloomberg report noted.


The decision follows an internal review of productivity and risk management. “After a period of experimentation, where we assess a variety of factors including productivity and risk management, we put them in the hands of those who are best suited to leverage them for the long term,” Goldman spokesperson Nick Carcaterra told Bloomberg.

Other banks, too, are looking at ways to limit access to internal information.

JPMorgan Chase is developing a controlled coding environment called Devspace for employees using Claude Code, Business Insider had reported. The environment is designed to isolate the tool from employees’ credentials, sensitive files and internal systems, while allowing the bank to monitor its activities.

Goldman not the only one

Goldman is not alone in reassessing employee access to external AI tools. Palantir, Nvidia and defence contractor Booz Allen Hamilton have restricted or considered restrictions on the use of Anthropic’s AI models over concerns around data security and intellectual property, as per reports in Reuters and The Information.

“Palantir has pressed Anthropic to provide irrevocable zero-data-retention guarantees before ⁠making its models available through Palantir’s software,” the reports said.

Nvidia, meanwhile, has limited its use of Anthropic’s models to less-sensitive tasks, such as open-source software projects. For sensitive internal work, including AI-driven supply-chain monitoring, it uses its own Nemotron models.

Booz Allen Hamilton has barred employees from using Anthropic’s models for work involving proprietary cybersecurity software it provides to customers. The company had previously used the technology to examine its systems for cybersecurity vulnerabilities.

“We worry a little bit that [Claude] might be learning from some of our code,” Bill Vass, Booz Allen's chief technology officer, told The Information, explaining the company's concerns around intellectual property.

Rising AI costs

Apart from security, the growing cost of AI tools is another reason for companies to restrict employee access.

Uber Technologies has capped monthly spending on AI coding tools such as Anthropic’s Claude Code and Cursor at $1,500 per employee, per tool, Bloomberg reported in June. The company introduced usage dashboards and has asked employees to seek permission to exceed their limits. This came after the firm exhausted its annual AI budget within months.

Walmart has also moved from unlimited access to a fixed allocation of tokens for its internally developed AI tool, Code Puppy, which employees use for tasks ranging from spreadsheets to presentations, Bloomberg reported.

Consulting firm Accenture, meanwhile, has introduced a monthly allocation of $29 in centrally funded Claude Enterprise credits for employees, with additional usage subject to separate budgets.