Unleash the power of AI safely! This article explores the security risks of using ChatGPT in your organisation and offers practical solutions to mitigate them. Learn how to leverage AI responsibly.
With the recent rise in AI programs, and the advancements that seem to be coming thick and fast, the question of whether AI can be harmful to your business is one that crops up again and again.
As security teams grapple with the new threats facing their companies, programs like ChatGPT can present them with challenges they may never have been confronted with before.
ChatGPT is used by employees around the world. It gives almost-instant answers to any questions they may have, breaking topics down, and sharing information in different formats for their benefit.
Its AI functionality allows it to write emails, suggest fresh ideas, create strategy documents and more, based on the information provided to it.
While it’s caused controversy around data protection in Italy (which has temporarily banned it) and plagiarism concerns, particularly in academic contexts, ChatGPT has become a useful tool for teams who might sometimes need an extra pair of hands.
Predominantly, from the information fed into it. Employees inputting sensitive data to ChatGPT may not be thinking of the consequences when they’re looking for a quick fix to a nagging problem.
According to recent research, sensitive data makes up 11% of what employees put into the system. That could include things like sensitive data such as PII and PHI.
Copying and pasting sensitive company documents into ChatGPT has quickly become a bad habit for employees who aren’t thinking of potential confidentiality issues and GDPR risks.
It’s not entirely clear how confidential information is handled from ChatGPT’s side. However, in their FAQ’s, they do state:
‘we are not able to delete specific prompts from your history. Please don’t share any sensitive information in your conversations.’
As employees use ChatGPT to be more efficient in their roles, they can intentionally or unintentionally share sensitive data with the tool. In so doing, they are feeding information into an LLM which uses data to learn from. The result is that ChatGPT could give this information back out to another user who is seeking answers on a particular issue.
ChatGPT itself says, 'It's crucial to be cautious and avoid sharing any sensitive, personally identifiable, or confidential information while interacting with AI models like ChatGPT. This includes information such as social security numbers, banking details, passwords, or any other sensitive data.
OpenAI, the organisation behind ChatGPT, has implemented measures to anonymise and protect user data. They have rules and protocols in place to ensure the confidentiality and privacy of user interactions. Nonetheless, it's always recommended to exercise caution and refrain from sharing sensitive information on public platforms, including AI chatbots.'
If an unauthorised user gains access to an individual's ChatGPT account, they can see a user's chat history, including any data shared with the AI tool. Users should ensure they have set strong passwords, and multi-factor authentication to minimise the chances of this happening.
Sensitive information shared with ChatGPT could be intercepted or compromised, giving malicious actors the opportunity to misuse personal information, share your organisation's intellectual property, or commit fraud.
As ChatGPT generates responses based on the data it has consumed, it can give out inaccurate or misleading information. If this is not addressed, and fact checked correctly, organisations may be at risk of reputational damage. In particular fields, such as finance and healthcare, where customer trust is essential, this may lead to a loss of trust with the business.
Bad actors can use ChatGPT to create email copy or messages that imitate a particular person, making employees susceptible to social engineering attacks. While hackers can sometimes be caught out by misspellings, or an inauthentic tone of voice, ChatGPT's conversational nature can lure individuals into thinking they are speaking to a genuine human - perhaps even one of their colleagues.
ChatGPT is designed to forget information once the conversation ends. However, there is a risk of data retention or improper handling of user data, which can result in ChatGPT creating a wide attack surface for bad actors to penetrate.
As with any tool, ChatGPT will likely have vulnerabilities that can be exploited. Whether the motive is to gain unauthorised access or extract sensitive data, malicious actors can exploit it to their advantage, leaving organisations at risk. Security teams should always be aware of employees using ChatGPT so they can ensure the relevant security measures such as patching and vulnerability assessments are undertaken.
It remains to be seen whether ChatGPT is a security risk right now, but there are always dangers associated with sharing sensitive data in a non-secure environment. You’ll always run the risk of data breaches and leakages, reputational damage, and financial losses.
There could well be a security risk in the future too. The National Cyber Security Centre has said that AI and Large Language Models (LLM) could help write malware in the future, and its natural language skills could also aid hackers in producing realistic phishing attacks.
For businesses, the amount of sensitive data that’s being put into ChatGPT could cause problems if hackers managed to infiltrate ChatGPT itself.
Chief Technology Officer at Metomic, Ben Van Enckevort, says,
"Whilst AI progress (ChatGPT et al) are an extremely exciting advancement in tech, managers should be aware of how their teams are using them - particularly with regard to the data that's been shared with these services. It's another factor security teams will need to take into consideration when they're thinking about their data security strategy. The rapid pace of change also means security professionals will need to be on the ball when it comes to keeping up with the latest threats."
It's important to remember that ChatGPT and other AI tools are currently hosted by third parties. In the future, companies may create their own AI systems but for now, any AI tool should be treated in the same way any other third party would be.
Educating your employees on the risks associated with ChatGPT is the best thing you can do to prevent sensitive information being shared with AI programs.
While some companies like JPMorgan Chase and Amazon have apparently banned employees from using ChatGPT and programs like it, you don’t need to go to those lengths to ensure they’re using them correctly.
Make sure everyone is aware that ChatGPT isn’t secure and despite its conversational nature, it shouldn’t be trusted with company secrets or customer information.
Setting up regular training sessions to update staff on the latest developments in AI is also a good idea. Your employees are one of your best defences against cybersecurity attacks so building up your human firewall is essential.
Our ChatGPT integration allows you to stay ahead of the game, shining a light on who is using the Generative AI tool, and what sensitive data they're putting into it.