Codex Security changelog: what's new each month
Every stable Codex Security release summarised by month: the highlights, new features, improvements, fixes and anything you need to act on. 5 months covered; the current month updates daily.
November 2026 · so far
7 releases: 0.8.3 → 0.10.0
November brought major enhancements to scan workflows, including batched Deep Scan reviews, enabled external scanner imports by default, and new resume capabilities for interrupted scans. Performance improvements and stricter security controls rounded out the month.
- Remove CODEX_SECURITY_ENABLE_EXTERNAL_IMPORTS from launch scripts; external imports now enabled by default and legacy value of 0 no longer disables them.
- Sparse SDK checkouts must include plugins/codex-security/mcp-app/src/deep-scan-settings.ts; upgrade or roll back bundled plugin as a unit.
- MCP registration scripts now require --all flag if relying on selecting all clients after automatic detection, or --agent NAME for single client selection.
Highlights
- External scanner imports now enabled by default, letting you preview and publish findings from supported external scanners.
- Resume interrupted Standard scans from saved checkpoints without losing usage or budget tracking.
- Deep Scan now runs independent Standard reviews in batches with worker collaboration and mutual verification.
- Batch scan budgets let you set a single cost limit across multiple repositories or components.
- Blocked patch results now capture detailed information about what stopped fixes and next steps.
New
- Resume capability for interrupted Standard scans using codex-security scans resume SCAN_ID.
- Batch scan budgets with --max-total-cost to share cost limits across repositories and retries.
- Hosted remediation now reports revision_changed when patch or verification ends at a different commit.
- Deep Scan budgets using --stop-after-no-new to count consecutive completed reviews adding no findings.
- Import ArmorCode SAST findings into the local workbench for analysis.
- MCP registration scripts with --all and --agent options for client selection.
- Blocked patch prompt asking what files changed, what stopped the fix, and next steps.
Improved
- Scan logs find sessions faster in large compressed histories.
- GitHub Action now includes scan scope limitations in job summary, including zero-finding scans.
- Standard scans now start with vulnerability audit and load host-specific instructions when needed.
- Deep Scan workers can share leads and verify each other's evidence in batched reviews.
- POSIX paths through symlinks now follow proper filesystem traversal.
- Plugin setup and CLI prompt deferral for faster startup.
- Native addons for Apple Silicon and Intel Macs included in npm package for Codex Desktop.
- Package verification rejects sparse tar entries for enhanced security.
Fixed
- Fixed surrogate pair splitting in truncated report text.
- Corrected output retention claims after scan failures.
- Fixed scan logs performance issues in large compressed histories.
- Addressed Linux Compose deployment security with updated seccomp profile.
October 2026
19 releases: 0.2.0 → 0.8.2
October brought significant expansions to Codex Security's scanning, remediation, and integration capabilities. The month saw multiple releases adding support for hosted workflows, external vendor imports, and enhanced recovery features.
- Findings service clients must send Content-Type: application/json to POST /v1/bulk/findings and POST /v1/dedupe-groups; missing or other content types return HTTP 400.
- Policy generation now writes threatmodel.md instead of THREAT_MODEL.md; integrations reading threat models must check SecurityPolicyDraft.threatModelPath for null.
- External imports now require opt-in with CODEX_SECURITY_ENABLE_EXTERNAL_IMPORTS=1 environment variable.
- Hosted verification uses new CLI verdict contract; remove imports of FindingVerificationCheck and replace reads of verification.checks with verification.status and verification.evidence.
Highlights
- Credential exposure detection now scans unused code and tests, distinguishing suspected exposures from placeholders.
- Investigation budgets let you set time limits for scans with --max-time, and multiple inference providers are now supported including Cheaper Inference, Amazon Bedrock, OpenRouter, and Fireworks.
- Hosted patching and verification allow integrations to generate patches and verify fixes through a JSON-RPC adapter.
- Import findings from Wiz, Snyk Open Source, Semgrep Code, Endor Labs, OpenGrep, GitHub Dependabot, and Amazon Inspector.
- Post-scan validation with --validate runs a supplemental validation pass and saves a separate report.
New
- Exposed credentials checking in source code, including unused code and tests, with distinction between suspected exposures and placeholders.
- Support for C++ headers (.hh, .hxx), server-rendered templates (EJS, ERB, PHTML), and Vyper source files in code selection.
- Shorter alias 'cs' for codex-security command.
- TypeScript SDK support for generating and verifying fixes with security.patch().
- External finding imports from Wiz, Snyk Open Source, Semgrep Code, Endor Labs, OpenGrep, GitHub Dependabot, and Amazon Inspector.
- Owner suggestions report with --owner-suggestions that preserves completed scan without assigning users.
- Investigation allowances with --max-time option (in minutes for CLI).
- Threat-model explorers to convert existing threat models into shareable interactive HTML explorers.
Improved
- Deep Scan efficiency improved by waiting for earlier results before starting additional reviews.
- Source selection now includes text files without recognized extensions, build and test files, and JSP and ASP.NET templates.
- Scan recovery preserves original knowledge context, model/provider settings, and selected paths.
- Terminal dashboard batches updates during busy scan replays and avoids rereading unchanged logs.
- Policy drafting now works when Codex is installed under the home directory.
- Standard and diff scans continue review in parent when delegation is unavailable.
- Deep Scan recovery retains unfinished checkpoints, worker coverage, and deadline recovery after cancellation.
- Finding review now shows critical findings first, followed by high, medium, and low severity.
Fixed
- Sandbox validation on Unix now checks that sandbox works before starting paid model calls, keeping completed results if follow-up work is cancelled.
- Bulk scans now preserve successful results and original scan failures when temporary-checkout cleanup fails.
- Reports now render malformed evidence ranges as a valid single line or just the path.
- Windows setup uses private workspaces in state directory to prevent elevated-sandbox permissions on TEMP from blocking scans.
- Scan directories now preserve trailing hyphens in repository names.
- Patch project-trust overrides are now serialized correctly for Codex.
September 2026
5 releases: 0.1.25 → 0.1.30
September brought significant improvements to scan management, finding deduplication, and patch verification workflows. The month introduced mock scanning for testing, custom severity classification, and enhanced support for multiple version control platforms.
- GitLab patch publication requires an installed and authenticated glab CLI.
- Interactive budget increases unavailable in CI, JSON/JSONL, headless, and verbose modes.
- Mock mode is opt-in and creates findings marked as synthetic in local history.
Highlights
- Mock scanning lets you generate synthetic results for testing without authentication or model calls.
- Custom severity classification uses your own rubrics to categorize findings, with assessments reusable for publication.
- Finding identities now persist across scans and comparisons, with automatic matching that speeds up lookups.
- GitLab patch publication support added alongside existing GitHub functionality, including self-hosted instances.
- Interactive budget increases available during scans when usage approaches limits.
New
- Mock mode generates synthetic Standard scan results with full report and export capabilities.
- Custom rubrics for finding severity classification with context-aware scoring.
- Automatic finding matching across scan history preserves confirmed identities and relationships.
- GitLab merge request creation for verified patches, supporting self-hosted GitLab.
- Context-aware cost estimate ranges reported during scans.
- Custom patch validation prompts for tailored verification workflows.
- GitHub Actions and Azure Pipelines workflow examples for Bedrock scans.
- Progress reporting while patching findings.
Improved
- Deduplication now uses separate screening and pair review with validated assignments.
- Large scan comparisons use bounded batches for better performance.
- Finding workflows preserve analytics settings across updates.
- Component scan progress reported in headless runs.
- Cost display made optional for flexible reporting.
- MCP server startup timeout increased to 120 seconds for bundled plugins.
- Scan settings now shared seamlessly between CLI and SDK.
- Session logs collected from Desktop and worker processes.
Fixed
- Replayed usage events with identical timestamps no longer counted in scan budgets.
- Invalid deduplication submissions now receive one corrective turn.
- Security fix verification skill now requires explicit invocation during other work.
- Local history database index migration applied automatically.
- JUnit parameterized names kept unique to prevent conflicts.
- Unavailable Daybreak access advisories skipped during scanning.
August 2026
18 releases: 0.1.6 → 0.1.24
August brought significant expansions to Codex Security's scanning and publication capabilities, with new inference providers, Linear integration, and cloud publishing support. The month focused on stability, removing artificial limits, and adding flexible workflows for validation and findings management.
- Removed authentication output and instruction limits.
- Removed remediation patch size limit.
- Removed local plugin and workbench output limits.
- Removed session event limits.
Highlights
- Added Linear integration to publish security findings and create patches directly from scan results.
- Introduced cloud publication to upload completed scans and findings to Codex Security Cloud.
- Launched configurable component scanning to split large repositories into explicit parts for targeted analysis.
- Added custom validation workflows and finding validation to recheck saved findings without rescanning.
- Expanded inference provider support with OpenRouter and Fireworks, plus configurable reasoning effort levels.
New
- Support for OpenRouter and Fireworks inference providers for security scans.
- Custom scan prompts and follow-up prompts to tailor scan behavior.
- Linear app integration to publish findings, create issues, and patch projects directly.
- Component scanning to divide repositories and scan each component separately.
- Custom validation workflows with setup, test, and cleanup steps.
- Finding validation to recheck saved findings without starting new scans.
- Preview findings service with storage, embeddings, and duplicate detection.
- CSV-based finding publication and GitHub code scanning alerts import.
Improved
- Unified Deep Scan and Standard scan phases for simpler scan workflows.
- Knowledge bases can now be shared across bulk scans.
- Verbose security scan diagnostics and token measurement across workers.
- Scan lifecycle and model metadata persistence for better tracking.
- Cost ceiling enforcement per repository with --max-cost option.
- Interactive finding review, patching, and optional pull request creation.
- Windows reliability improved for credential handling and case-insensitive CODEXHOME.
- Scan session details now available for live observation of Deep Scan progress.
Fixed
- Resolved Deep Scan worker recovery after failures and coordinator restarts.
- Fixed multiscan supervisor locks to be safe across container restarts.
- Corrected knowledge-base input validation during dry-run scans.
- Resolved Windows credential ACL verification stability and performance issues.
- Fixed Linear publication recovery and credential reconciliation for cloud uploads.
- Patched high-severity js-yaml advisory in dependencies.
July 2026
6 releases: 0.1.4 → 0.1.5
July saw Codex Security reach its initial public release with a TypeScript SDK and Node support, alongside multiple patch releases stabilizing core security scanning, authentication, and Windows compatibility.
Highlights
- Launched public TypeScript SDK and Node.js workflows for developers to integrate security scanning.
- Fixed critical authentication conflicts and improved API key precedence handling.
- Added reasoning effort controls to the security CLI for customizable scan behavior.
- Stabilized Windows support with path normalization and platform-safe handling.
- Implemented automatic scan comparison matching and npm/GitHub release automation.
New
- TypeScript public SDK for programmatic access to security scanning.
- Reasoning effort controls in the CLI for adjusting scan depth and analysis.
- Automatic scan comparison matching to streamline result tracking.
- False-positive feedback sync with update notices for continuous improvement.
- Automated npm and GitHub release workflows with verification.
Improved
- Enhanced authentication handling with clearer API-key precedence and conflict resolution.
- Scan targeting now preserved when security scan targets or configurations change.
- Windows compatibility improved with canonical path casing normalization.
- Scan output now kept private through completion and export operations.
- CLI help and deep-scan configuration now properly documented.
- Git hook and authentication tests made Windows-compatible.
- Personal accounts now included in bulk-scan discovery.
- Registry authentication improved for container manifest digests and image attestation.
Fixed
- Resolved scan authentication conflicts and state errors affecting multi-account setups.
- Fixed actual security scan failures to surface properly instead of being masked.
- Corrected malformed security scan findings recovery to prevent data loss.
- Fixed npm release tarball smoke tests and package lifecycle handling.
- Resolved broken pipes when reading container manifest digests.
- Made historical GitHub release backfills idempotent to prevent duplication.
Summaries are written automatically from the official release notes (full changelog ↗); check the original notes before relying on a detail. Codex Security: pricing, features and alternatives · All changelogs