From the AWS edition of October 31, 2026
AWS publishes a locked-down configuration for running the ChatGPT desktop app and Codex against Amazon Bedrock in AWS GovCloud (US-West)
Amazon Web Services (with OpenAI ChatGPT desktop app / Codex) · Government cloud; developer and knowledge-work agents · United States (AWS GovCloud US-West)
- Publisher
- AWS Artificial Intelligence Blog
- Original publication
- 2026-10-30
- Source retrieved
- 2026-11-01
What happened
An AWS technical walkthrough (Oct 30) says the ChatGPT desktop app, with its Work and Codex components, can send model inference to Amazon Bedrock Mantle in us-gov-west-1 using AWS IAM identities. The guide sets an administrator requirements.toml that forces API-only sign-in, disables browser and computer use, memories, plugins, MCP servers and telemetry, and allows only the Bedrock GovCloud endpoint while denying OpenAI and ChatGPT domains. AWS says processing stays in the selected Region, billing is per token with no seat licenses, and zero data retention is supported. It states plainly that following the guide does not authorize a workload; the agency's own system security plan and authorization process still apply. ChatGPT workspace rules and roles do not govern this setup, so device policy and AWS permissions are the main controls.
Why it matters
State IT shops and public universities that already hold AWS GovCloud accounts get a documented path to pilot an agentic coding and research assistant without sending prompts to a commercial SaaS tenant. That matters where CJIS, IRS 1075 or state data-residency rules have kept such tools off agency desktops. Relevance is bounded: the guide targets organizations whose security architecture requires GovCloud inference, and most local governments and K-12 districts do not operate GovCloud accounts.
Evidence and measured results
Vendor technical how-to with full configuration files, troubleshooting and rollback steps. No customer deployment, usage figure, cost comparison or independent security test is reported.
Limitations and uncertainty
Vendor guidance only; no deployment evidence. Requires a desktop app version that supports Bedrock in GovCloud, which AWS tells readers to confirm with OpenAI. Disabling web, browser, plugins and memories limits the tool's usefulness relative to commercial versions. Authorization remains with the customer.
Put this evidence to work
Lighthouse Advisory interpretation, grounded in this source. Enriched 2026-11-01; this does not change the original publication date. Each role follows the same checklist so takeaways compare across sources.
Sales
Role takeaway
- Customer problem
- Agencies want AI coding and research assistants but cannot send prompts or code to commercial SaaS tenants under their data rules.
- Stakeholders
- State or university CIO
- CISO and ATO authorizing official
- Application development manager
- Budget office
- Discovery questions
- Do you already operate AWS GovCloud accounts with Bedrock enabled?
- Which data classifications would developers handle with the assistant?
- Can your endpoint management enforce admin policy files on developer workstations?
- How would you budget metered per-token spend?
- Value hypothesis
- A tightly restricted assistant inside the existing GovCloud boundary may let a small developer cohort test AI-assisted coding without a new SaaS authorization; value is unproven until the pilot measures it.
- Potential engagement
- A four-to-six-week controlled pilot design and security review for one development team, ending in a go/no-go memo.
- Unsupported claims
- That the configuration is authorized for any workload
- Productivity or cost savings
- That all ChatGPT features work in GovCloud
Pre-sales engineering
Role takeaway
- Fit
- Agencies with GovCloud accounts and managed developer workstations; poor fit for organizations without GovCloud.
- Architecture and integration
- Client on managed devices calls the Responses API on Bedrock Mantle in us-gov-west-1 using IAM or IAM Identity Center credentials; app-level network allow-list blocks other endpoints.
- Prerequisites
- Approved GovCloud account and IAM role with model access
- Desktop app release that supports Bedrock in GovCloud
- Endpoint management to deploy requirements.toml before first launch
- Constraints
- Only us-gov-west-1 is covered in the walkthrough
- Browser, computer use, plugins, MCP and memories disabled
- ChatGPT workspace roles do not apply
- Security
- Protect credential files from agent writes, keep secrets out of TOML, and confirm the effective identity because a Bedrock bearer token overrides the profile.
- Proof of value
- Verify the app shows amazon-bedrock as provider and cannot reach OpenAI domains
- Confirm disabled capabilities remain off after restart
- Test that an unapproved API key cannot change the identity used
Delivery
Role takeaway
- Work
- Stand up a controlled pilot with policy files, IAM roles, monitoring and rollback for one developer team.
- Dependencies
- Agency ATO or security review
- OpenAI app release compatibility
- Bedrock model access in GovCloud
- Ownership
- Agency CISO owns the policy baseline; the application development manager owns pilot operations and budget.
- Skills and adoption
- Developers need training on reviewing agent-generated code and on the restricted feature set.
- Governance checkpoints
- Security review of configuration before first device
- Midpoint review of usage logs and spend
- Go/no-go decision against written stop conditions
- Acceptance criteria
- Proposed: 100% of pilot devices pass the provider and network checks
- Proposed: per-token spend stays within an agreed monthly cap
- Proposed: participating developers report whether the tool helped completed tasks
- Risks
- Configuration drift on unmanaged devices
- Credential misuse through precedence of bearer tokens
- Restricted tool proves too limited to justify effort
Implementation considerations
Lighthouse Advisory interpretation across the operating dimensions a public-sector buyer must settle before this evidence becomes a design. Each note answers the question under its heading for this specific source.
Architecture and integration
What must connect, and where does the AI sit in the workflow?
Inference routes through the Responses API on Bedrock Mantle in us-gov-west-1; the client runs on managed workstations with a named AWS profile or Bedrock API key (the bearer token takes precedence over the profile). Network allow-listing at the app level is the main containment boundary.
Governance
Who approves, reviews and stays accountable for outcomes?
Because ChatGPT workspace roles do not apply, governance shifts to endpoint management, IAM role design and the agency's ATO process. Agencies need an owner for requirements.toml and a change process for app updates, which the guide disables at startup.
Security and privacy
What data, permissions and controls need testing?
Credentials must stay out of TOML files and be protected from the agent's own workspace writes. The precedence of a bearer token over the profile can silently change which identity is used, a failure the guide lists in troubleshooting.
Accessibility and workforce
Who is affected, and what skills or accommodations follow?
Codex shifts developer work toward reviewing agent-produced code; agencies need review norms and training before broad use. No accessibility findings are reported.
Procurement
What should contracts, pricing and exit terms secure?
Per-token billing with no seat licences changes budgeting from per-user subscriptions to metered consumption, which needs cost guardrails and chargeback in state budgeting cycles.
Operating model
Which teams own the service once it runs?
Rollout is framed as a controlled pilot with explicit stop and rollback steps, which fits a security-led operating model rather than a self-service rollout.
Publication history
- 2026-10-31AWS · Issue 037 resources
Stable resource ID: aws-chatgpt-desktop-codex-bedrock-govcloud-guide-20261030