/Tech1d ago

Google integrates native computer use into Gemini 3.5 Flash with built-in safeguards against prompt injection

Story Overview

Google has embedded a computer-use tool directly into Gemini 3.5 Flash through the Gemini API, letting the model observe screens via screenshots and generate actions such as clicks, typing, and navigation across browser, mobile, and desktop setups in a repeating loop.

1632.1K158731200.7K

#684

Original post

Philipp Schmid@_philschmid#1404inTech

We're launching computer use in Gemini 3.5 Flash today. Give your agent a screen and a goal, it figures out the actions.

- Supports browser, mobile (phones), and desktop as environments - Integrated safeguards, user confirmation, auto-stop on prompt injection - Additional training against prompt injection - Native tool build into the main Gemini 3.5 Flash model.

I've been using this to test our docs. I told id "audit this (url) page, run the code snippets and find issues" and it navigates through, screenshots, and comes back with a report.

9:25 AM · Jun 24, 2026 · 85.9K Views

Developer Impact

Client-side setup shapes real-world use

Developers must supply their own execution environment and sandbox, with examples using Playwright and Docker; the feature supports Python, JavaScript, and REST calls but leaves runtime security and cost details to each implementation.

Safety Focus

Safety layers stay configurable rather than automatic

Built-in policies cover areas like financial actions and data changes, while prompt-injection detection is opt-in only; high-risk steps still require explicit human confirmation under the recommended guidelines.

Sentiment

Positive users praise Gemini 3.5 Flash's native computer use for smooth cross-platform agent tasks, while negative users dismiss the Flash model and demand Gemini 3.5 Pro instead.

Pos

71.4%

Neg

28.6%

22 comments with sentiment.

Cluster Engagement

Digg Deeper

No Digg Deeper questions have been answered for this story yet.