The ChatGPT Desktop App Now Runs on Amazon Bedrock in AWS GovCloud (US)
By MetaPatchAI News Team
Primary source: AWS Machine Learning Blog: "Configure the ChatGPT desktop app with Amazon Bedrock in AWS GovCloud (US)" by Virgil Ennes and Sean Phuphanich, published Oct 30, 2026 (in AWS's RSS feed at 1:27 PM ET). Checked Oct 30, 2026 ~8:10 PM ET.
Oct 30, 2026: Government agencies and contractors that can't send data to OpenAI's own servers now have an AWS-sanctioned way to run ChatGPT's desktop tools. AWS's new post opens: "The ChatGPT desktop app is now available on Amazon Bedrock in AWS GovCloud (US)." Model requests go through Amazon Bedrock in the us-gov-west-1 Region instead of through OpenAI's servers.
The post is a technical walkthrough, not a press release, but it's the clearest official confirmation we've seen of this setup.
What you get
According to AWS, the desktop app provides access to OpenAI's frontier models "through two components":
- Work: "completes complex, multi-step tasks and delivers review-ready results"
- Codex: OpenAI's coding agent "for developing, testing, and reviewing software," also usable through the Codex CLI and IDE integrations with VS Code, JetBrains and Xcode
Inference is routed through the Responses API on Amazon Bedrock, using an endpoint AWS calls Amazon Bedrock Mantle.
Why GovCloud matters
AWS says the walkthrough is for "organizations whose security architecture requires model inference to use AWS GovCloud (US) identities, endpoints, and Regional controls." Per the post:
- Data stays in-Region: "processing stays within the Amazon Bedrock AWS Region that you select"
- Pay per token: "no seat licenses and no per-developer commitments"
- Zero data retention (ZDR) is supported: in that mode, AWS says prompts and responses "are processed without being written to durable storage by AWS or shared with model providers"
- Access control through AWS IAM and existing GovCloud network controls
One important caveat: because the app talks to AWS directly in this setup, AWS says "ChatGPT workspace rules and user roles don't control it." Security comes from device policies and AWS permissions.
How the locked-down setup works
The walkthrough has three steps:
- Authentication. Use an approved GovCloud account and IAM identity. The app accepts either an Amazon Bedrock API key (the
AWS_BEARER_TOKEN_BEDROCKenvironment variable, which takes precedence) or the standard AWS SDK credential chain, including AWS SSO through IAM Identity Center. AWS says to keep credentials and tokens out of the config files. - Two config files. An admin-controlled
requirements.tomlsets limits users can't override. A userconfig.tomlpoints the app at theamazon-bedrockprovider. Both have to exist before the first launch. - Launch and verify. Confirm the app shows
amazon-bedrockas its model provider, either in the Codex CLI/statusview or in the app's settings.
AWS's baseline requirements.toml is strict. Among other things it:
- allows only API-mode sign-in
- turns off browser use, computer use, image generation, memories, plugins and MCP servers
- allows the Bedrock GovCloud endpoint and explicitly blocks OpenAI domains such as
chatgpt.comandapi.openai.com
The fine print
AWS includes several warnings worth repeating:
- "Using AWS GovCloud (US) and following this guide do not, by themselves, authorize your workload." Agencies still have to review the model, services, endpoint and data flows against their own security plan.
- If the app reports the GovCloud endpoint as unsupported, AWS says to confirm the exact app release, endpoint and model with OpenAI, because "changing the URL alone does not establish support."
- AWS's sample config doesn't give a real model ID; set
model = "<your-approved-model-id>"and use the model ID your GovCloud account has approved. AWS doesn't name which OpenAI models are approved in GovCloud. - The example AWS profile is named
codex-il5. AWS doesn't explain the name, and the post doesn't state any DoD Impact Level or FedRAMP authorization for this setup, so don't assume one.
Who should care
- Agency IT and security teams who've been asked for "ChatGPT, but compliant"
- Defense and federal contractors already running workloads in GovCloud
- Developers in those environments who want Codex without sending code to OpenAI-hosted endpoints
What's confirmed / not confirmed
Confirmed (AWS blog, Oct 30, 2026):
- The ChatGPT desktop app (Work + Codex) can use Amazon Bedrock in AWS GovCloud (US-West)
- Inference goes through Bedrock Mantle's Responses API in us-gov-west-1; per-token pricing; ZDR supported
- Setup uses API sign-in, admin requirements.toml, user config.toml and an allow-list that blocks OpenAI domains
Not confirmed: - Which specific OpenAI models are available in GovCloud, and their pricing - Any FedRAMP or DoD Impact Level authorization for this configuration - A matching announcement from OpenAI. We found only the AWS post. - Whether the same app setup is supported in AWS GovCloud (US-East). The post covers US-West only.
Related
Up: AI
