Rapid growth of artificial intelligence (AI) means platforms are handling increasing amounts of data, but how safe is that data and what are the consequences for AI companies?
To help us understand the question, we asked Copilot for its thoughts. The answers demonstrated why AI companies need to be vigilant and get the right insurance cover.
Why we asked Copilot
We get asked a lot whether insurance covers data breaches in AI platforms. It’s a valid question given how many companies are creating AI products, and how many others are relying on those products.
In theory, the answer is yes, but only if you have the right AI insurance, typically including cyber insurance and supported by professional indemnity.
But much also depends on what AI platform you’re creating.
Most small AI firms and platforms harness the datasets and large language models (LLMs) of other companies and tech giants.
Those systems are normally designed with safeguards to protect personally identifiable data, but some may store it. Which is why we decided to ask one of the major AI platforms what it thought.
What we asked Copilot
We posed a simple question: does Copilot store sensitive data that I input into it?
Copilot flattered us with its initial response – “That’s a very sensible question to ask”– before saying:
“Your inputs aren’t used in a way that identifies you personally, but they may be processed and used in limited ways to improve the service—depending on how the system is configured.”
That seemed a little vague.
What Copilot’s full response told us about data safety
ICopilot went on to say: “systems and data may be used in anonymous or aggregated form” and that “while safeguards are in place, no AI system should be treated as a secure vault for sensitive or regulated data”.
Finally, and most importantly it suggested AI systems can “in rare cases . . . regurgitate memorised fragments from training data” which could include personal information.
That told us everything: it’s never advisable to put personal data into any AI system.
Protections in self-learning models
Lots of AI firms harness models from other larger tech firms, with many adapting them for their own needs.
Where they are tailored to ‘self-learn’, they can store questions customers ask, to inform future answers.
If a system’s capabilities allow additional self-learning or creation of bespoke AI solutions, it must be designed with appropriate safeguards to protect personal data inadvertently inputted into the system.
Protections in self-learning models
Getting the right AI insurance is a must, but there are also things you can do to protect your business, including:
- Analyse and assess, what personal data can be accessed from your AI models or solutions.
- Ensure safeguards are in place or put in place to ensure no personal data can be accessed by those using your AI platform.
- Ensure users and/or customers know that entering personal data comes with risk.
- Try to design systems that delete personal data immediately or store it so that it can’t be co-mingled with other data in the model.
- Commit significant resource to implementing the most robust cybersecurity processes if your AI solutions and data sets include personal data.
Trust Copilot, get covered
Whatever you think of global AI platforms like Copilot, take note of how it responded to our question. The best way to protect your AI business against data claims is to get the right insurance.
Speak to our friendly team of experts today.
