Interest in AI surged after the release of ChatGPT in November last year and tech optimists were predicting that these new, more sophisticated large language models (LLMs) would soon be used by every business, speeding up processes and liberating us from drudge work.
But only a few months later, several large businesses moved to ban the use of the AI chatbot in the workplace, with the likes of JPMorgan Chase, Apple and Accenture all restricting their staff’s use of ChatGPT.
The primary concern for these businesses was security. Many feared that the prompts employees typed into the AI’s interface would be used to improve the model or that OpenAI – the creators of ChatGPT – would have access to their queries.
Other issues with the current staple of AI products include the lack of ability to customise answers; the fact that ChatGPT is only trained on information up to 2021; hallucinations, where an AI confidently submits a completely incorrect answer; and trustworthiness, which is still a concern for 46% of business leaders.
What is ChatGPT Enterprise?
OpenAI hopes that its latest iteration of ChatGPT, which is designed specifically for businesses, will solve many of these problems. ChatGPT Enterprise, it promises, offers greater security, faster responses and more advanced capabilities, including new customisation options.
“This enables businesses to roll out a generative AI foundation without having to invest in substantial new infrastructure or skills from the outset,” explains Rowan Curran, a senior analyst for Forrester who specialises in generative AI. “Many people are accustomed to using ChatGPT in their work, whether it was permitted by their organisation or not, and this now gives them a way to do so safely and securely.”
Curran predicts that the security changes and new features, such as the ability to extend its use cases via plugins, may lead to greater adoption of OpenAI’s products at an enterprise level.
If we want to create the market leading advisory models, our proprietary data is the key IP
Australian graphic design platform Canva was among a handful of businesses that were granted access to ChatGPT Enterprise ahead of its public release. Its head of AI products, Danny Wu, says that the business is now using the Enterprise version across numerous departments, including engineering, data analytics and finance. “It’s become a true enabler of productivity, with the dependable security and data privacy controls we need,” he adds.
Further down the line, OpenAI plans to allow Enterprise users to train the AI on their own data so that they receive specialised responses, and to release more function-specific tools and capabilities.
PwC has also been experimenting with ChatGPT Enterprise for the past few months. Its UK CTO, Bivek Sharma, says: “All the issues we had in terms of security and being able to use ChatGPT in a B2B environment have now been resolved.” The unlimited message cap and the quicker speed of responses have also proved useful, he adds.
The accountancy firm has ambitions to develop its own AI models for its advisory services. “If we want to create the market-leading advisory models, our proprietary data is the key IP,” Sharma explains. “The real brain power has already gone into building these large language models; what we’re now doing is training it in specific use cases and adding our own datasets.”
Is ChatGPT Enterprise worth it?
Emma Haywood is a legal consultant at IP and technology specialist Bloomworks Legal. She doesn’t think that the Enterprise version of ChatGPT will solve every issue that businesses had with the platform. If, for example, companies use AI to create content, then ChatGPT still has the potential to copy other people’s work. This can leave companies open to accusations of IP infringement. “Courts around the world are still grappling with these questions,” she says, “So it isn’t necessarily safe to assume that the content that you create using ChatGPT Enterprise won’t create risks for your business further down the line.”
ChatGPT Enterprise’s compliance with SOC 2 – a widely recognised security standard – and OpenAI’s commitment not to use customer data to train its models are also “a step in the right direction”, according to Haywood. But companies must still be considerate of the information they enter on the platform.
You could end up with a much bigger regulatory burden within your own business even though you didn’t technically create the tool
Use of personal information will continue to be regulated under GDPR and client data is often protected contractually, so plugging it into a third-party platform might not be permitted. Similarly, companies could lose any protections pertaining to trade secrets if they willingly enter them into a chatbot. “Don’t just blindly assume that, because you have these new features, it’s completely safe to type everything in,” Haywood says.
ChatGPT Enterprise is not the only AI product on the market that offers these capabilities. PwC is working with a range of AI platforms, including Microsoft’s Azure AI and Google’s generative AI Bard. “We are watching how the various platforms are evolving, and making sure that we are pioneering on all fronts, to ensure we are using the best models for the job,” Sharma says.
When deciding which platform is best for your business’s needs, cost, performance and security will be top of mind. However, he advises that the ability to access the model, train it on your data and develop your own models from it is “the really important bit”.
What impact will AI regulation have on business’s adoption?
AI regulation, which is rapidly being drawn up by the EU, the US and the UK, also has the potential to create problems for businesses further down the line. The ability of businesses to tailor AI models to their needs may blur the boundaries between user and provider in the eyes of some regulatory bodies.
Haywood explains: “If you customise ChatGPT Enterprise to the extent that you’re not just a user, you’re then the provider of the model. That means you could end up with a much bigger regulatory burden within your own business, even though you didn’t technically create the tool. We don’t yet have a clear definition of where the line is.”
This creates another level of uncertainty for business users of these products and shows that, while some issues surrounding corporate use of AI tools are being resolved, new ones continue to be created.
