Why Companies are Banning ChatGPT
The hidden danger of AI chatbots
This month Apple joined other tech and financial service giants, such as Verizon, Samsung, JP Morgan, and Amazon, in restricting work-related use of ChatGPT due to data security concerns.
Chatbots and Parrots
For large companies such as Verizon and Apple, allowing employee access to ChatGPT is like allowing parrots in the office.
Similar to how a parrot mimics what it hears, ChatGPT generates responses based on what it’s trained on, which includes user inputs. In turn, when an employee inputs sensitive information to the chatbot, there’s a chance ChatGPT parrots that information to someone else.
This possibility is why ChatGPT prompts users to not share sensitive information with the chatbot — there’s no guarantee it will remain confidential.
Though the chance of data leaking for any given input is small, when thousands of employees use the chatbot the probability of a catastrophic-leak increases. It’s as if the company brought in thousands of parrots instead of just one.
There is also concern that smart actors may be able to reverse-engineer prompts to ChatGPT to uncover confidential data (almost like reverse training the parrot).
When I asked ChatGPT to help me reverse-prompt confidential info, it gave me some compelling techniques:
Don’t try these at home, or ChatGPT may ban you too.
Bans Everywhere
Given these security concerns, it is likely that other companies, especially those entrusted with sensitive customer data, will also enforce restrictions or bans on the usage of ChatGPT, if they haven’t already.
The last thing a company like Morgan Stanley wants is a summer intern leaking the details of an undisclosed merger because they asked ChatGPT to write their email.
Other AI chatbots may also be up for ban.
When Samsung engineers uploaded sensitive meeting notes and internal code to ChatGPT, the company banned all generative AI chatbots, including Google Bard.
This willingness to ban an established company like Google suggests that it’s not OpenAI which companies are skeptical of, but the underlying generative AI technology.
The randomness of generative AI models makes them untrustworthy for sensitive data, and companies don’t want to take the risk of a leak.
ChatGPT Trick
There is one safeguard to help prevent ChatGPT from inadvertently leaking sensitive information.
Users can choose for their chats to be marked as unsaved in ChatGPT’s settings, which results in ChatGPT deleting the chat after 30 days. This safeguard prevents user inputs from being incorporated in ChatGPT’s long-term model.
There is no guarantee however that this feature prevents leakage during the 30 day period before deletion. Moreover, it is unlikely that companies can rely on employees to enable this setting consistently.
In the long-term, OpenAI and Google will likely need to incorporate other privacy protocols if they want their chatbot allowed in major corporations.
In the meantime, the sign on the wall is clear — parrots are banned from the office.
Some Final Thoughts
If you’re reading this article, your world is likely going to be disrupted by AI.
We don’t know what the changes will be, but they’re certainly coming.
There’s a chance we all become human can-openers for robot overlords, collecting dust in a cupboard soon. There’s also a chance that AI unleashes ridiculous, ungodly levels of human talent, and we make the Renaissance look like an afternoon preschool session.
No one really knows.
At Byte-Sized Insights, we believe everyone should be informed of AI’s advancement, regardless if it’s good, bad, or sideways. Each week we post Medium articles on progress made in the technology and business of AI.
They’re simple, quick, and designed for you, the people.
Follow using the link below.
Click here to follow Byte-Sized Insights before the AI overlords make us can-openers
