Data Privacy & Security
AI models rely on new data to ensure their corpus can evolve and remain up-to-date. This is done via re-training. Many AI’s are re-trained with data entered within user prompts. Data collection can be turned off on some AI models. However, many people are unaware of this and will obliviously enter personal and sensitive information without understanding the potential repercussions. Examples of information you should avoid entering into an AI prompt include (but are not limited to):
- Banking information
- Identification documents (passport, driving license, national insurance number)
- Legal documents (contracts, court documents, employment contracts)
Why is this a risk?
Models can ‘memorise’ parts of their training data. Especially unique entries. As a result, they can output information very similar to it’s source, resulting in a potential leak of sensitive information.
How can I avoid complications?
When entering information into a prompt, ensure that any sensitive information is redacted. This can include:
- Personal identifiers: full name, address, phone number, email
- Financial details: banking information, credit card numbers
- Authentication data: passwords, security information
- Health records or legal information