Use Gemini 3.5 Flash computer use only after sandbox checks
Google made computer use a built-in Gemini 3.5 Flash tool for browser, mobile, and desktop agents, but teams still need sandboxing, confirmation prompts, and prompt-injection checks before production use.
Google has confirmed computer use as a built-in tool in Gemini 3.5 Flash. Confidence level: confirmed public preview. Developers can now build agents that see, reason, and act across browser, mobile, and desktop environments through the Gemini API and Gemini Enterprise Agent Platform.
Caption: Google's Gemini 3.5 Flash artwork for the computer use announcement. Source: Google Blog.
What changed
Google's June 24, 2026 announcement moves computer use from a standalone Gemini 2.5 computer-use model into Gemini 3.5 Flash as a native tool. The Gemini API release notes list the feature as a public preview and name simplified actions with intents, browser, mobile, and desktop support, configurable safety policies, and advanced prompt-injection detection.
The practical shift is workflow design. Teams can combine computer use with Gemini 3.5 Flash's other tool paths instead of routing screen-control tasks through a separate model. Google positions the feature for long-horizon automation, continuous software testing, and professional-app workflows.
| Capability | Best fit | Access | Status | Caveat |
|---|---|---|---|---|
| Gemini 3.5 Flash computer use | Browser, mobile, and desktop agent workflows | Gemini API and Gemini Enterprise Agent Platform | Public preview | Requires sandboxing and workflow tests |
| Sensitive-action confirmation | Form submits, purchases, deletions, irreversible actions | Optional enterprise safeguard | Available with the feature | Must be configured deliberately |
| Prompt-injection stop behavior | Webpage or document content that tries to steer the agent | Optional enterprise safeguard | Available with the feature | Detection is not a complete security boundary |
Why this is early
This is official, but still early for production planning. Google says developers and enterprises can start using the feature now, while the Gemini API changelog labels it public preview rather than a settled general-availability default.
The Next Web independently confirms the same shift and notes that Google has not published full updated benchmark detail for the integrated Flash tool versus the older standalone model. Treat the release as usable for pilots, not as permission to run unsupervised agents against sensitive systems.
Key takeaways
- Gemini 3.5 Flash now has built-in computer use for screen-facing agents.
- Access runs through the Gemini API and Gemini Enterprise Agent Platform.
- Google calls out browser, mobile, and desktop environments, not only web pages.
- The feature includes optional safeguards for sensitive actions and indirect prompt injection.
- Teams still need sandboxing, least-privilege accounts, audit logs, and human approval for risky steps.
Availability and access
The Gemini API release notes list computer use in Gemini 3.5 Flash as a public preview. Google says developers can start through the Gemini API, while enterprises can use the Gemini Enterprise Agent Platform.
Pricing and limits need account-level verification. The feature may use Gemini 3.5 Flash pricing, but real cost depends on screenshot loops, tool calls, retries, and how often confirmation prompts interrupt a workflow. Check your own API project, quota, region, and enterprise platform setup before designing a rollout.
Practical LinkLoot angle
Computer use becomes useful when a process still depends on screens: regression testing, admin panels, legacy dashboards, form-heavy workflows, or professional apps without clean APIs. Gemini 3.5 Flash may lower the friction because the same model can reason and act, but the security model matters more than the demo.
Start with one contained workflow. Use a non-production account, block access to secrets, limit network scope, record actions, and require confirmation before the agent submits, deletes, buys, exports, or changes permissions. For broader tool selection, compare agent stacks in LinkLoot's AI workflow automation guide.
What to verify before you act
- Confirm that computer use appears in your Gemini API project or Gemini Enterprise Agent Platform environment.
- Review the current Gemini docs for model IDs, preview terms, quotas, regions, and pricing.
- Test the optional sensitive-action confirmation flow before exposing real accounts.
- Run prompt-injection probes against webpages, documents, and dashboards the agent will read.
- Keep browser, mobile, and desktop tasks in sandboxes with low-privilege credentials and auditable logs.
Source check
Confirmed by:
- Google's announcement confirms computer use is now built into Gemini 3.5 Flash and available through the Gemini API and Gemini Enterprise Agent Platform.
- Google's Gemini API release notes also list the June 24, 2026 public preview, but this generic changelog URL is not included in the publication source list because LinkLoot has already used it for a separate Gemini API update.
Independent context:
- The Next Web independently reports the same native-tool shift, the replacement of the standalone Gemini 2.5 computer-use path, and the enterprise safety framing.
Google says developers and enterprises can start using it through the Gemini API and Gemini Enterprise Agent Platform, with the release notes labeling it public preview.