Colab's free AI coding help comes with an 18-month data window

The setup

Colab is Google's free, browser-based Jupyter notebook environment, widely used for teaching and quick prototyping because it needs no local Python install, and it now ships generative AI features built into that same free tier. Google's own Colaboratory FAQ states the AI features, reachable through a Gemini icon in the notebook footer and inline autocompletions, are enabled by default for eligible users: an account must belong to someone 18 or older and sit in a supported locale. No separate signup is described; eligibility is the only gate the FAQ names for the free product.

What the documents show

The FAQ describes the features in plain terms: generating and transforming code, fixing errors, running an autonomous data-science agent, and analyzing a user's data from a prompt. Google's Colab product page markets the same capability as Gemini generating and running code to answer a natural-language question. Neither page publishes an accuracy figure; both describe capability, not measured performance. On data handling, the FAQ is direct: using the generative AI features means Google collects the prompt, related code, the generated output, usage information and any feedback given, and uses that data to provide and improve Google's products, including enterprise products such as Google Cloud.

The friction

The same passage states human reviewers may read, annotate and process those prompts, outputs and feedback, and warns against including sensitive or personal information in either. It adds that this data is stored in a way where Google can no longer tell who provided it, so a later deletion request cannot be fulfilled, and it is retained for up to eighteen months. That is a real cost for anyone using the free tier for coursework, a client's dataset, or unpublished research code, disclosed by the FAQ rather than inferred by this note. The FAQ mentions Colab Enterprise, a separate paid product, only in passing, without stating whether the same terms apply there.

What changed in the work

For students, researchers and anyone prototyping without a machine to configure, the FAQ supports a real change: AI code generation and error-fixing that used to require a paid or locally installed assistant now sits inside a notebook people were already opening for free. What it does not support is treating that access as privacy-equivalent to a paid product; the 18-month retention and human-review disclosure are specific to the free-tier terms, and a team handling anything sensitive would need to check a paid plan's separate terms before assuming otherwise.

• Does any notebook opened in free-tier Colab touch client, student or unpublished data that should not sit in a stored prompt for 18 months?
• Is the account meeting the age and locale eligibility the FAQ names, and would that change on an institutional account?
• Would a paid or Enterprise tier's terms actually differ here, and has that been checked rather than assumed?

Google's own FAQ makes the free tier's AI features easy to verify and use, enabled by default rather than requiring setup, but the same document is where the retention and human-review terms live, the trade a free notebook environment asks users to accept without a separate consent step.