OpenAI DevDay 2026 adds Codex security and new APIs
The newest TechCrunch report adds that Sam Altman framed the Decisions API as a limited preview constraining Luna to predefined choices, and that TypeSafe CEO Diogo Almeida likened it to Jev, citing a QueryStory demo at about $2.94 versus $372 per agent action and a prior Hugging Face incident. Earlier coverage had only described Luna and, in The Decoder, a ten-times-faster GPT-6 claim.
OpenAI’s DevDay 2026 launched Codex Security Cloud and new APIs; later reporting cast Decisions as a limited Luna classifier akin to Jev.
At DevDay 2026 in San Francisco, OpenAI announced Codex Security Cloud, a research-preview appsec service for ChatGPT Pro, Business, Enterprise, and Edu that scans entire GitHub repositories on demand, on a schedule, or on new commits, builds editable threat models, validates candidate vulnerabilities in isolated environments, deduplicates findings, and prepares patches that become draft pull requests only after human review, with no automatic code changes. The cloud tier includes Daybreak Blue cyber-capable models by default for authorized defensive work, role-based admin controls, and continued scanning when a developer’s laptop is closed. OpenAI also made Codex cloud environments persistent and reusable across computer, phone, and cloud, refreshed the Codex CLI with voice control and an /agents view, and added ChatGPT desktop review of GitHub pull requests and GitLab merge requests. API updates include an Agents API with Computer Use, multi-agent features, Tool Search, and Context Compaction, AWS Bedrock Managed Agents for OpenAI agents running on AWS, and, per The Decoder, an Ultrafast Codex tier at 300 tokens per second priced at six times standard rates ($60/$300 per million tokens for GPT-6 Astra). Sources differ on the Decisions API: The Decoder said a GPT-6 Luna version returns answers ten times faster, while later TechCrunch reporting said Sam Altman described a limited preview that constrains Luna to predefined choices and that TypeSafe CEO Diogo Almeida compared it to TypeSafe’s Jev, citing a QueryStory Jev demo at about $2.94 versus $372 per agent action.
- At DevDay 2026 in San Francisco, OpenAI announced Codex Security Cloud as a research preview for ChatGPT Pro, Business, Enterprise, and Edu users, with role-based admin controls.
- It scans entire GitHub repositories on demand, on a schedule, or on new commits; builds editable threat models; validates candidate vulnerabilities in isolated environments; deduplicates findings; and drafts patches that become pull…
- The cloud tier bundles Daybreak Blue cyber-capable models by default for authorized defensive work such as vulnerability discovery, malware analysis, and incident response, and keeps scanning when a laptop is closed.
- Codex cloud environments are now persistent and reusable from computer, phone, or cloud with shared team settings; the CLI adds voice control, an /agents view, and a cleaner terminal; the ChatGPT desktop app reviews GitHub pull requests…
- The Agents API gains Computer Use, multi-agent features, Tool Search, and Context Compaction; AWS Bedrock Managed Agents can build OpenAI agents that run entirely on AWS.
- The Decoder reported an Ultrafast Codex tier at 300 tokens per second, six times standard API pricing ($60/$300 per million tokens for GPT-6 Astra), and a Decisions API on GPT-6 Luna that returns answers ten times faster.
- A later TechCrunch report said Sam Altman described the Decisions API as a limited preview that constrains Luna to predefined choices; TypeSafe CEO Diogo Almeida, a former OpenAI engineer, likened it to TypeSafe’s Jev.
- A QueryStory demo of Jev, not of OpenAI’s API, checked each agent action for about $2.94 versus $372 with a frontier model; advocates said that approach could have caught a prior Hugging Face agent incident.
Coverage
oldest first · one row per article · UTC- · Sep 29, 2026
The DecoderFirst report
OpenAI expands Codex and its API at DevDay with security scans, a Decisions API, and Ultrafast
OpenAI's DevDay 2026 adds Codex Security Cloud repo scanning, Computer Use in the Agents API, a Decisions API, and an Ultrafast tier.
- · Sep 29, 2026
TechCrunch · AI
OpenAI gives Codex reusable cloud environments that work across devices
OpenAI made Codex cloud environments reusable across devices and added CLI, review, and security tools.