Commonwealth Bank of Australia restricts staff from using ChatGPT

Australia’s largest lender, Commonwealth Bank of Australia, has restricted its employees from using popular artificial-intelligence chatbot ChatGPT, instead setting up and operating its own in-house system.

CBA revised its Group AI policy in the past three months, clarifying that staff couldn’t use ChatGPT on CBA devices nor take CBA information that isn’t publicly available and put it into external chatbots, Chief Data and Analytics Officer Andrew McMullan told The Wall Street Journal.

“The update to our policy was clear on our position for the use of ChatGPT or similar chatbots that exist,” he said.

The lender made the decision as part of its regular review of AI developments, and McMullan said that in May he wrote to all senior CBA leaders to reiterate policy changes.

ChatGPT has grown in popularity since OpenAI released it in November, with people using it to automate tasks at work and school. But organizations are blocking the use of ChatGPT, with some citing concerns that it could lose control of customer information.

Apple in May restricted the use of ChatGPT and other external AI tools for some employees. Similar mandates came from companies such as JPMorgan Chase and Verizon earlier this year.

An OpenAI spokeswoman pointed to an April announcement where the company introduced users’ ability to turn off their chat history, which the company said would block the ability to train the AI model on that data.

McMullan said CBA is taking an “internal approach” to using generative AI, where technical staff use a ChatGPT-like tool called CommBank Gen.ai Studio. The tool developed alongside H2O.ai, its Silicon Valley-based partner, leverages 49 global large language models, which have been brought on to CBA platforms.

“We have been using large language models and multi-modal models on our data, on our platform, on our environment, coded and supervised by CBA employees so that we can understand everything that we’re doing,” said McMullan.

“Our AI policy is unbelievably important so that we can explain and understand everything that we’re doing with AI…we’ve been very thoughtful in making sure that what we build on our platforms and on our systems.”

Businesses have been turning to in-house solutions and a growing ecosystem of vendors that aim to protect sensitive company data that could be fed into new AI tools.

H2O.ai Chief Executive Sri Ambati said every organization needs to own its large language models and generative pre-trained transformers, in the same way that it needs to own its brand, data, algorithms and models.

“Open source generative AI is bringing exponential transformation to our customers by democratizing value from AI while preserving data, code and content ownership,” he said.