OpenAI must explain action taken to stop AI hacking Australians’ private data, chair of federal inquiry says

Data privacy concerns are causing companies to reconsider their choice of AI vendors, fearing that these companies could potentially misuse customer intellectual property to enhance their models. A prominent example of this controversy occurred in June when Anthropic introduced a 30-day retention policy for chats with its Fable and Mythos models.

This decision sparked significant backlash from businesses, who expressed their frustration over the matter. In response, influential figures such as Microsoft CEO Satya Nadella and Palantir CEO Alex Karp voiced their concerns, warning companies against relying on AI models from frontier AI companies, as they risk having their proprietary knowledge leaked or used to build competing products.

Despite their statements that they do not train their models on enterprise data, data security experts remain worried about the potential for subtle data extraction methods that can improve future AI models without directly training on customer data. This has led to a growing desire among businesses to diversify their AI usage and reduce their reliance on leading AI companies.

The term "sovereign AI" has gained traction in recent months, referring to a company's ability to control its AI stack, including owning the hardware, setting up proprietary cloud infrastructure, and using open source models in their own cloud environments.

However, implementing sovereign AI comes with challenges. It requires a higher level of technical expertise and may not provide access to the latest capabilities offered by AI giants like OpenAI, Anthropic, and Google. For instance, running open source AI models may limit businesses' access to advanced functionalities such as coding or financial analysis.

Nonetheless, many companies are exploring hybrid models that combine closed models and open source alternatives. This approach allows businesses to maintain partnerships with leading AI companies while also utilizing open source models for specific use cases, ensuring they retain control over their data and intellectual property.