Antigravity has begun showing users a new in-app notice warning that AI coding agents carry security limitations, including the risk of data exfiltration and possible code execution.
Antigravity · Security notice
Antigravity warns that AI coding agents carry security limits
A new in-app notice flags data exfiltration and possible code execution as general risks of AI agents that read files, run commands, and act on a user's behalf. It advises limiting sensitive data and checking every action.
Why the agent's reach matters
Each capability widens what the agent can touch
Reads files
Access to project content
→
Runs commands
Can execute instructions
→
Acts for the user
Takes actions without each step being typed
Risks named in the notice
Data exfiltration
Sensitive information leaves a system without authorization.
Code execution
The agent runs instructions it should not.
What users are advised to do
Limit sensitive data
Avoid having the tool process highly sensitive data.
Verify every action
Review what the agent does before accepting results (human-in-the-loop).
Still unconfirmed
The notice does not say whether technical safeguards are coming
When the notice first appeared
Whether it applies to all users
Whether the developer plans product or permission changes
Until more information is provided, treat the warning as guidance to limit sensitive data and review agent actions closely.
Continue reading The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.
Already purchased? Sign in ✓ Signed in — this article isn’t included in your current plan.Unlocking the full article...