OpenAI's coding agent moved into AWS. The compliance unlock matters more than the model.
Jump to a section
See the story at a glance
GPT-5.5, GPT-5.4, and Codex went generally available on Amazon Bedrock on June 1, 2026. One month of limited preview, then GA. The model lineup itself isn't the story — these have been available directly from OpenAI for months. What changed: every API call now inherits AWS's compliance stack.
IAM permissions, VPC and PrivateLink isolation, KMS encryption, CloudTrail audit logging. For organizations that have already passed SOC 2 and FedRAMP reviews on their existing AWS stack, OpenAI models on Bedrock don't require a new security review. They inherit the one you already have.
That's the thing that was blocking enterprise adoption. Not performance. Not pricing. The security paperwork.
Codex now has more than 5 million weekly users — up from 4 million at the April preview launch. A 25% jump in one month. Those 5 million are mostly connecting directly through OpenAI. The Bedrock path is the on-ramp for the regulated industries that weren't in that count before.
Pricing matches OpenAI's first-party rates, which is table stakes. No AWS-native discount. But usage counts toward Enterprise Discount Program commitments, which matters if your org has active EDP targets it's trying to hit.
OpenAI also confirmed that Daybreak — their AI-powered security initiative launched May 11 with partners including Akamai, Cisco, Cloudflare, and CrowdStrike — will be the next major capability arriving on Bedrock. No date given.
Source spread
- OpenAI — Frontier models and Codex on AWS — hype. Official launch post; leads with developer adoption numbers and governance framing.
- AWS Blog — Get started with OpenAI on Bedrock — builder. Technical integration details, user count, regional availability.
- Help Net Security — OpenAI models on AWS — builder. Regulated-industry framing; governance controls specifics.
- About Amazon — Bedrock OpenAI models — hype. Business partnership angle, AWS executive quotes.
Pros & cons
What's genuinely useful:
- The compliance unlock is real. IAM + VPC + CloudTrail on OpenAI inference is a meaningful change for banks, health systems, and government contractors who couldn't run OpenAI models without a separate security review. The review is now implicit in your existing Bedrock approval.
- EDP credit counting is a real procurement accelerant. If your org has AWS commit targets, Codex and GPT-5.5 spend now moves the needle on them.
- Codex's IDE integrations (VS Code, JetBrains, Xcode) route through Bedrock when you configure the Bedrock endpoint. No client plugin changes required — just swap the API base.
- Anthropic's Claude models are already on Bedrock. This means you can now A/B test Claude and OpenAI models through the same compliance-approved platform with the same logging.
What deserves scrutiny:
- This is a distribution move, not a capability move. Codex and GPT-5.5 don't get better because they run on Bedrock. If you were already on OpenAI's direct API and happy with performance, nothing improves.
- No AWS-native discount. The EDP benefit only applies if you have active commitments. Most smaller orgs don't.
- GitHub Copilot has had enterprise governance controls on Azure longer. If your org is Microsoft-native, Copilot may already have the approvals Codex would need to replicate, and switching has a migration cost.
| Feature | Bedrock | OpenAI direct API |
|---|---|---|
| Governance (IAM, VPC, KMS, CloudTrail) | Inherits existing AWS controls | Separate security review required |
| Pricing | OpenAI first-party rates | OpenAI first-party rates |
| AWS EDP credit counting | Yes | No |
| IDE integrations (VS Code, JetBrains, Xcode) | Yes — configure endpoint only | Yes — default |
| Data residency | Matches AWS region config | OpenAI terms |
| Compliance framework (SOC 2 / FedRAMP) | Inherits from existing Bedrock approval | Separate review process |
- Check your existing Bedrock approval first. If Claude is already approved in your org's Bedrock environment, the security review for OpenAI models likely follows the same framework — same governance controls, same audit scope.
- EDP math: If you have active AWS EDP targets, add expected Codex and GPT-5.5 spend to your calculation. Real dollars for orgs hitting commit milestones.
- Client setup is a one-liner: Set
CODEX_API_BASEto your Bedrock endpoint or configure it in Codex App settings. IDE integrations (VS Code, JetBrains, Xcode) pick it up without plugin changes. - Run your workload against both. You can now run Anthropic and OpenAI models through the same Bedrock endpoint with identical logging and access controls. Do the comparison on your actual tasks before picking a primary — synthetic benchmarks are not your codebase.
- Wait one quarter on Daybreak. If you're evaluating security scanning tools, Daybreak lands on Bedrock next. Evaluating Glasswing/Claude Security vs. Daybreak in the same compliance context will give you a much cleaner comparison than the current split-vendor setup.
Further reading
- OpenAI — Frontier models and Codex on AWS — official launch post
- AWS Blog — Get started with OpenAI GPT-5.5, GPT-5.4, and Codex on Bedrock — technical integration guide
- OpenAI — Daybreak cybersecurity initiative — the next major Bedrock addition from OpenAI
- OpenAI — Codex overview — product background
- AWS — Amazon Bedrock compliance certifications — SOC 2, FedRAMP, HIPAA coverage
Liked this? Get the weekly digest.
Free. Monday mornings. The week's stories, synthesized. Unsubscribe anytime.
Your take
How'd I do on this one?
What did I miss?
Tell Samwise (and Sam).
Disagree with the take? Spotted a fact I got wrong? Have context I should have included? Drop it here. Anonymous unless you leave an email.
Keep following the thread
A little more to explore.
30,000 AI-generated merges a day broke git. Anthropic built something else.
By September 2026, AI agents were merging over 30,000 code changes into Anthropic's codebase each day — up from 2,500 in May. Git stopped scaling. So seven engineers spent five months building Scion, an in-house version control system designed for the merge volumes that AI-era development actually produces.
robots.txt gets a shared AI training opt-out. Finally, one that might work.
OpenAI, Google, Microsoft, and Anthropic agreed on October 30 to honor a single robots.txt token — __usage_ai-training — for opting out of AI training. One directive, four crawlers, no more maintaining separate blocks for GPTBot, ClaudeBot, and Google-Extended.
Anthropic's 280-million-install standard just became everybody's. That's the right call.
Anthropic donated Agent Skills — the SKILL.md open standard for giving AI agents specialized workflows — to the Linux Foundation's Agentic AI Foundation. With 70+ platforms integrated and 280 million installs, it's following the same vendor-neutral governance path MCP took earlier this year.