OpenAI pops an enterprise sticker on ChatGPT to give big biz some peace of mind
Here's what you actually get for this VIP level. And how is Microsoft happy with this?
OpenAI launched ChatGPT Enterprise on Monday, a tier of the text-generating chatbot focused on alleviating concerns about privacy and other fears business customers may have. What does enterprise-level access actually get you?
Let's get some easy points out of the way first. In announcing the enterprise tier on Monday, OpenAI said "all conversations are encrypted in transit and at rest." Bear in mind, all conversations with ChatGPT, from the free tier up to enterprise, via the API or bot interface, are encrypted in transit, using HTTPS, or specifically, TLS 1.2.
Regarding the encryption of conversations at rest, OpenAI said it uses the standard strong algorithm AES-256 to secure enterprise customer data in storage. That said, a spokesperson told The Register "our non-enterprise chats are encrypted as well."
In an FAQ, OpenAI added that even for enterprise customers, some of the lab's staff can unlock and access these encrypted business conversations when necessary:
Within your organization, only end users can view their conversations. Workspace admins have control over workspaces and access.
Authorized OpenAI employees will only ever access your data for the purposes of resolving incidents, recovering end user conversations with your explicit permission, or where required by applicable law.
Next, OpenAI is keen to stress that enterprise-tier customers' data and conversations are not incorporated into ChatGPT's model, declaring: "We do not train on your business data or conversations, and our models don’t learn from your usage."
But we note that, according to the lab, OpenAI does not use data submitted through the API to train or improve ChatGPT for all tiers, enterprise or not. And for non-API, consumer-level nattering with the chat bot, data provided may be used to train the model – but you can opt out of that.
So far, to us at least, the enterprise tier appears to give you the same level of encryption as non-enterprise; offers the same level of data privacy via the API as for non-enterprise; and for the chat bot interface, it doesn't train the model from conversations by default whereas consumer-grade bot users have to opt out of that.
OpenAI needs to push the privacy and cryptography facets of its enterprise offering because companies including Apple, Amazon, Samsung, JPMorgan Chase, and Bank of America have banned employees from using ChatGPT. Big business fears that asking the chatbot to analyze internal docs or code, or hold conversations with customers, may lead to sensitive or proprietary data being inadvertently absorbed into the model to be later regurgitated and leaked to third parties.
OpenAI would like to wash away those worries. "You own and control your business data in ChatGPT Enterprise," the lab stressed in its marketing bumf. We're reminded too that OpenAI warns that ChatGPT is not perfect, and may produce inaccurate results however confident it sounds.
So what is apparently special to the enterprise tier? Administrative controls, for one thing: a console to manage an organization's ChatGPT users in bulk; support for single-sign-on authentication; verification of users' domain; and an analytics dashboard for monitoring bot usage.
Another major point is that ChatGPT Enterprise customers are promised unlimited access to a GPT-4-powered ChatGPT that runs up to twice as fast on OpenAI's servers versus non-enterprise GPT-4 usage; and access to its Code Interpreter data-analysis tools.
Enterprise customers can use 32,000-token context windows (tokens are used to measure and limit the size of input data; one token represents about four characters of English-language words), though we see on OpenAI's pricing page that non-enterprise GPT-4 customers can get that size context window, too, if they're willing to pay.
Users within an enterprise org can also share conversation templates with colleagues to collaborate and complete tasks by building on prompts and sharing outputs.
As one point of comparison, individual ChatGPT Plus users are only allowed to send up to 50 messages every three hours to OpenAI's GPT-4 model.
- Apple becomes the latest company to ban ChatGPT for internal use
- Stop OpenAI training its models on your chats by turning off history
- Small custom AI models are cheap to train and can keep data private, says startup
ChatGPT Plus costs $20 per month, whereas large companies will have to negotiate with OpenAI's sales teams for the cost of ChatGPT Enterprise. Outfits such as PwC, Estée Lauder, Block, Zapier, and others are said to be customers. More than 80 per cent of Fortune 500 companies that have registered ChatGPT accounts have reportedly adopted some of the chat bot's functions since launch, we're told, though we'll see if they go full enterprise.
OpenAI is planning to add more features for enterprise users in the future to support customization. It says it wants to build new tools that will make it easier for ChatGPT to securely access company data and connect to other applications, as well as developing new capabilities to fulfill tasks performed by specific roles such as data analysts, marketers, or customer service.
"We look forward to sharing an even more detailed roadmap with prospective customers and continuing to evolve ChatGPT Enterprise based on your feedback. We're onboarding as many enterprises as we can over the next few weeks," it said.
This enterprise offer is interesting in that it pits OpenAI against its biggest champion, Microsoft, which has poured billions into the super lab. The IT giant thus already has various enterprise-grade OpenAI-powered chat-bot offerings in its Azure cloud. CTOs will have a choice: either go with Microsoft, and make use of any Azure and Office integrations already available for OpenAI's ChatGPT; or go straight to OpenAI and hope it's enterprise ready. ®