The Google Trap Software Developers & Founders are Walking Right In, and Never Know

If you have ever used an AI to code, refine your code, or in any manner whatsoever, fully agentic or not, you should be worried, and part of those I wrote this for.
If you're also building a startup or software that involves code and you have a developer working for you on it, you should be worried, too, and like the category above, I wrote this for you.
So, I spent the better part of my weekend and Thursday (weeks ago) reading Google's privacy policy, specifically as it relates to the company's Agentic coding tool, Antigravity.
Now, Antigravity is of two kinds — the Agentic-only Antigravity, or simply Antigravity, and the Antigravity IDE. Both are IDEs, of course, in the technical sense of it, but the latter allows you the flexibility of writing your code yourself with the help of an army of AI models when you NEED them.
The former, as you may have guessed, does not allow that latitude; just send in the prompt and let the agent or your preferred model burn tokens out.
Now, regardless of the specific Antigravity variant, Google's policy still applies. And that policy is where issues begin to emerge.
But before then, let's look at something similar, but with a confusing nature — telemetry.
Google says in both Antigravity apps that toggling off the app’s telemetry prevents it from sending data. What data is that? They gave some hints, but omitted a strong keyword or key phrase or key words (whatever you chose to call it)
That term is ‘Interactions Data’. This phrase is key because it is used in Google policies.
By Google's standards, regarding the use of its AI products, interaction data is the data you, the user, generate from using its AI products. That includes your conversations, your codebase, the code the AI generates, and the code you reject.
Basically, every code submitted to its AI coding platform is received by Google, but that's not even the full issue.
The Google visibility trap
Everyone collects data, isn't it?
The problem is what that data is used for.
Google says usage of its Antigravity platform is subject to three policies, including the Universal Terms.
Find it here in verbatim:
The following documents govern your use of the Service:
the Google Terms of Service (the “Universal Terms”);
these Google Antigravity Additional Terms of Service (the “Google Antigravity Terms”); and
the Google Privacy Policy (the “Privacy Policy”).
Now, without going through all these terms, which is great that you do, the company made it easier to understand how your data is used by explicitly stating that it uses your data to train its models.

Note: Google was very clear that you read these policies.
Have a read, verbatim as usual:
When you use the Service, we record and store your user data, interaction data pertaining to your usage of the Service, related metadata connected to the Service, and any feedback you provide (“Interactions”). Such data may be aggregated over multiple users, and will be collected only when you have the Service running. You will have the option to delete your Interactions. If you would like to request that your Interactions be deleted, you can email antigravity-support@google.com. Note that such Interactions will be used in accordance with the terms of this Agreement unless and until you request deletion in accordance with the previous sentence
The problem, however, is that Google was not explicit about what it meant by “interactions.”
Does it involve any of what I mentioned earlier? I can’t tell.
But inference from Google's refusal to answer direct questions clarifying that, their reputation for data and privacy handling, and from mere English on what Interactions means in the context of using a platform like Antigravity suggests something is oddly off.
Google sees your code. But it doesn't end there. It trains its models on your code and allows its employees and contractors to view your code.
Now, I don't know what privacy agreements Google staffers have with the company or the security clearance required to access people's code, but still, there's something off about the platform's choice of path.
And while one may argue that Google already has more information about its users from other services, or that there are millions of Antigravity users with potentially trillions of lines of code, so the probability of picking one's codebase is significantly low… I still wouldn’t fully buy.
Yes, I very much agree with you on that, but the fact that there are Reddit posts on these, online questions asked, and that people build stuff they don't want an AI ingesting so it can spit out part or all of it sometime later says a lot, doesn't it?
There is an exception, though. Enterprise users subscribed to Enterprise Cloud or Workplace plans don't have their code treated this way.
Here it is in verbatim:
If you are accessing the Service through Gemini Enterprise (Google Cloud), Gemini Enterprise for Business or a Google Workspace subscription on the Google Cloud Pre-GA Offering Terms, then you are subject to the terms of use accepted or signed by your administrator applicable to the Service and the terms below do not apply to you.
Google recognizes that enterprise codes must be treated with utmost privacy, and I laud the company for that.
My argument on this…
The entire reason for this, however, is that not many enterprise users even recognize that these protections, or the absence of them in some contexts, exist. And some who know don't use that protection because it's more expensive.
Plus, there is a sizable number of individuals building ‘private’ stuff, or stuff that should be protected, while doing so on a closed-open system like Google Antigravity.
Startup founders and outsourced developers fill these categories.
A second question that I’d leave open, largely because the legislation around it is nuanced at the moment, but I believe falls more in favor with end users based on where I'm coming from, is the issue of copyright to code.
If an AI agent generates part or all of your code, who holds the copyright for that code?
Now, copyright laws treat computer code like literary works, and as such, once you write code, the copyright for that code is automatically yours.
But with AI-generated codes, did you really write the code, and as such, are you capable of holding the copyright to that code? Can you sue a company like Google for storing and using ‘your code’ to create a similar product because you used its Antigravity platform to code.
Remember now that you supplied the prompt, and because the AI agent can't generate a code out of thin air, your prompt is the foundation, but you didn't write it. Can this be akin to outsourcing coding to a freelancer and using a binding contract to require them to relinquish any copyright they have in that code to you or your company? Well, there is no binding contract with Google that relinquishes copyright from AI to Vibe Coder. But does the AI have the copyright in the first place?
Can an AI claim copyright for a code it generated using your prompts, refinement, and reiterative guidance? This is an open question, but I believe an AI model — or the platform that owns it — can't, and should not, claim copyright to it.
If a military general takes the blame and praise for the success or failure of a war, even when they never fought in the front lines, then the user (coding general) should take the credit or the opposite of a generated code, just as they should be prosecuted if they abuse the AI agent to generate malicious code.
—-----
PS:
Oops… you may ask: should AI companies not be blamed for users using their platforms to cause harm? If they fail to put adequate safeguards and lines in place, then yes, they should be blamed. But if a user uses their creative genius to devise a novel way to bypass existing strong guardrails, then the onus is on the platform to help law enforcement prosecute that user. I could write another article on this, but I need to study, and this very last paragraph is an add-on to the one right above it.
The military general's logic can also be flipped.
