Public AI Tools Create Risk You Can't See
Employees already use AI at work, often with tools your IT team never approved, on infrastructure you don't control.
Last updatedof privacy and security professionals admit entering non-public company information into GenAI tools
of organizations have had data leak through employees sharing sensitive information with AI tools
of breached organizations had a shadow-AI incident, up from 20% a year earlier. Those breaches averaged $5.39M
Same Prompt In. Very Different Journey Out.
Every AI tool starts the same way: someone types a prompt and hits send. What happens in the next second is the whole difference between a public tool and a private one.
Consumer and Unapproved AI Tools
Your prompt leaves your network the moment you hit send. It's processed on shared infrastructure, kept under someone else's policy, and sometimes passed to sub-processors you never see or approve.
Inceptus Reeve
Your prompt is processed on your own box, under your control. No model provider sees it, nothing outside your box learns from it, and nothing is kept beyond the retention you set.
Enterprise plans from the big AI vendors close some of these gaps by contract. Reeve closes them by keeping the model and the data together, under your control. See how it works →
What the Big AI Providers’ Own Terms Say
Nobody’s terms say “your data isn’t private.” Many say they can train on it, have people read it, keep it for years or use it for ads. These are the free and personal plans your employees are most likely using on their own.
“Please don’t enter confidential information that you wouldn’t want a reviewer to see or Google to use to improve our services.”
Chats a person has reviewed are kept for up to three years, even after you delete your activity.
“When you use our services for individuals, such as ChatGPT and Codex, we may use your content to train our models.”
Training is on until the user switches it off, and feedback can bring a whole conversation back into training.
“We will train new models using data from Free, Pro, and Max accounts when this setting is on.”
With training on, chats are kept for five years instead of 30 days.
Since December 16, 2025, Meta uses people’s chats with its AI to personalize content and ads.
Users of Meta AI can’t opt out.
Its privacy policy says information it collects, including prompts and uploaded files, is kept on servers in the People’s Republic of China.
No fixed retention period is stated.
In 2025, a federal court ordered OpenAI to preserve consumer ChatGPT chats, including deleted ones, and later to produce 20 million de-identified chat logs in a lawsuit.
Enterprise customers were carved out. Consumer accounts were not.
Business and enterprise plans from these providers generally don’t train on your data by default. The bigger risk is the personal accounts your people use without approval, and that even a good contract still leaves your data on someone else’s infrastructure. [Quoted as of October 2026. Re-check each against the live page, and have counsel review, before publishing.]
Public AI Risk: Common Questions
Can deleted AI chats be used in a lawsuit?
They can. In 2025, a federal court in New York Times v. OpenAI ordered OpenAI to preserve consumer ChatGPT conversations, including deleted ones, and later to produce 20 million de-identified chat logs. Enterprise and zero-retention customers were carved out. Anything typed into a third-party AI tool can become a record someone else holds.
Do people read what employees type into AI tools?
Sometimes. Google says a subset of Gemini chats are reviewed by human reviewers, including trained service providers, and that reviewed chats are kept for up to three years, even after you delete your activity. Google asks users not to enter confidential information they wouldn’t want a reviewer to see.
Are enterprise AI plans safe enough for sensitive data?
They’re much safer than consumer plans: most don’t train on your data by default and come with stronger contracts. They still keep your data on the vendor’s infrastructure, under its retention rules, with sub-processors and exposure to legal demands. For CUI, patient records, privileged files or trade secrets, keeping the model inside your own network removes those dependencies.