Image: Google
If you plan to point Gemini at your code to hunt for security holes, it may be worth thinking twice before you use your main Google account. Developer Artem Russakovskii says Google sent him several policy warnings tied to Gemini, and his best guess is that a popular bug hunting prompt set them off.
Just got several alarming, scary, and ridiculous notifications from Google. Apparently, I used Gemini in some way that violates Google policies, and if I do it again, I may lose access to unspecified services and content.
— Artem Russakovskii (@ArtemR) October 26, 2026
The emails said his account had been used in a way that breaks Google's rules. The listed service was Gemini, and the stated issue was "trying to interfere with or disrupt Google's services or infrastructure, or attempting to use Google services for disrupting others' infrastructure or services." Google told him his access had not been restricted, but warned that repeat violations could cost him access to specific services. Russakovskii said the email gave no details about what triggered it, and that he was driving when the alerts arrived.
What he thinks triggered it
In a follow up post, Russakovskii said the only thing he could think of was installing Google Antigravity for the first time the day before to run a security review of his company's code. He used a prompt called Black October v2, shared on X by Franz Bruckhoff, with Gemini 4 Argon as the model.
Black October Prompt v2:
— Franz Bruckhoff (@franzbruckhoff) October 23, 2026
Run an exhaustive static security audit using a hardened Recon, Hunter, Critic, and Ledger workflow
The prompt asks an AI agent to run an exhaustive static audit of a code repository using isolated subagents, based on an open source skill Cloudflare published on GitHub. It tells the model to hunt for auth bypasses, leaked secrets, supply chain flaws and bugs similar to the recent PHP2Shell issue, then propose fixes. Russakovskii said Gemini ran out of Pro credits and stopped before it got far.
His complaint is about how this was handled. He wrote that Google could have simply refused the prompt, but instead let it run and then sent a vague warning a day later. Another user said they got the same warning after asking only about the PHP2Shell vulnerability. StatsWire picked up the story and urged developers using Gemini for code scans to be careful.
If you're using Gemini 4 Argon inside Google Antigravity to scan your own code for security bugs, be VERY careful.
— StatsWire (@StatsWire) October 26, 2026
A developer says he ran a security scan using AI agents.
Hours later, Google flagged his account for potentially disrupting its services.
None of this is confirmed. Google has not said why the warning was sent, and Russakovskii himself calls the Antigravity session a guess. Google's generative AI prohibited use policy does bar content that helps with "interference with, or disruption to Google's or others' infrastructure or services," though it also allows exceptions for educational and scientific uses. The Antigravity terms say users must not interfere with or disrupt the service and are responsible for what their AI agents do. Google has not commented publicly at the time of this writing.
What you can do
Read the warning email closely and check your account notices before you do anything else. If access is restricted later, Google says you'll get a separate email with a way to appeal.
Think about running heavy security audits on a separate Google account, not the one that holds your email, photos and Drive files. Several people in the replies said they plan to do just that.
Keep prompts focused on code you own, and say so plainly. Aggressive wording about finding exploits may look like attack planning to automated filters. Google has also said in a Gemini CLI post that flagged Antigravity accounts can be reinstated through a form, but a second violation may lead to a permanent ban.
Comments