Context-heavy vulnerability review
Investigate flaws whose impact depends on application trust boundaries, data flows and surrounding business logic.
Independent tool overview
Codex Security at a glance
Codex Security is OpenAI's application-security agent for building repository-specific threat models, exploring attack paths, validating suspected vulnerabilities and proposing reviewable patches.
Visit the official Codex Security site ↗
- Best for
- Engineering and application-security teams reviewing GitHub repositories
- Availability
- Research preview for ChatGPT Pro, Business, Enterprise and Edu
- Repository support
- GitHub repositories at the time of review
- Core workflow
- Threat model, discovery, sandbox validation, patch and revalidation
- Code changes
- Proposed for human review; not applied automatically
- Separate price
- No standalone Codex Security price publicly listed
Overview
What Codex Security is
Codex Security is a research-preview application-security product inside Codex. It connects to selected GitHub repositories, analyzes the code and commit history, and builds an editable threat model describing entry points, trust boundaries, sensitive data and high-impact paths. That context guides its search for vulnerabilities instead of relying only on static signatures.
For a suspected issue, the agent can attempt reproduction in an isolated validation environment, record an attack path and proof-of-concept evidence, and propose a minimal patch. The patch does not modify the repository automatically; a human must review it and can turn it into a pull request. This workflow can add valuable security depth, but it remains one control in a broader program—not a replacement for dependency scanning, secrets detection, dynamic testing, architecture review or professional penetration testing.
Use cases
Who Codex Security is best for
The strongest fit depends on the job you need the product to complete, not the size of its feature list.
Historical repository scans
Analyze current code plus commit history to look for existing vulnerabilities and relevant architectural context.
Attack-path validation
Give security reviewers reproduction evidence and explicit assumptions instead of only a rule match and severity label.
Remediation assistance
Generate a minimal patch proposal and revalidate the issue after a reviewed fix is merged.
Security teams piloting agentic review
Start with a small repository set and dedicated reviewers before integrating results into a wider vulnerability program.
Capabilities
Core Codex Security features
GitHub repository connection
Enable selected GitHub repositories for initial and ongoing analysis through Codex Security.
Editable threat model
Generate a project-specific model of entry points, trust boundaries, sensitive assets and high-impact code paths that teams can inspect and correct.
Commit-history analysis
Scan commits in reverse chronological order to build context and search for vulnerabilities already present in repository history.
Agentic vulnerability discovery
Use language-model reasoning, large context, tools and test-time compute to explore realistic code and attack paths.
Sandbox validation
Attempt to reproduce potential issues in an isolated environment before surfacing them as validated findings.
Attack-path analysis
Show how attacker-controlled input may travel from an entry point to a sensitive outcome, including likelihood, impact and assumptions.
Proof-of-concept evidence
Record reproduction results and execution details that reviewers can use to judge exploitability and priority.
Minimal patch proposals
Generate a focused remediation suggestion intended to address the root cause without silently changing the repository.
Pull-request workflow
Allow a reviewed proposal to become a pull request in the team's normal code-review process.
Post-fix revalidation
Re-run validation after a confirmed issue is patched and merged to help close the remediation loop.
Enterprise access controls
Enterprise and Edu admins can gate Codex Cloud, Codex Security use and scan administration through roles or SCIM-synced groups.
Process
How the Codex Security workflow works
Step 1
Choose a controlled pilot
Start with a small set of repositories and reviewers, using lower-risk code first if GitHub Cloud access is new to the organization.
Step 2
Configure access
Authorize only required repositories and roles, review data handling, and separate scan administration from ordinary finding access.
Step 3
Correct the threat model
Add deployment assumptions, real trust boundaries, sensitive assets, authentication context and excluded test-only behavior.
Step 4
Review validated findings
Inspect the attack path, reproduction details, assumptions, severity and affected code instead of accepting the label alone.
Step 5
Test the patch independently
Review the diff, run unit, integration, regression and security tests, and confirm the fix does not break intended behavior.
Step 6
Merge, revalidate and learn
Use the normal pull-request and release process, revalidate the fix, then update the threat model and security backlog from reviewer feedback.
Cost
Codex Security pricing and free plan
OpenAI does not publish a standalone Codex Security fee or per-scan rate. The research preview is available through eligible ChatGPT subscriptions, subject to rollout, workspace controls and usage limits. Plan prices below are the broader ChatGPT subscription prices, not a dedicated security-scanning quote.
ChatGPT Pro
$200/month
Individual-plan eligibility for the Codex Security research preview.
- Codex Security availability is confirmed in the current help article
- Usage and flexible-credit availability can vary
- Consumer data controls apply
- Not a team security-governance workspace
ChatGPT Business Standard
$20/user/month annual or $25 monthly
A self-serve team workspace with Codex access and business data protections.
- Two-seat minimum
- Included usage limits with optional workspace credits
- No training on business data by default
- Confirm Codex Security entitlement and limits in the workspace
ChatGPT Business Premium
$100/user/month annual or $125 monthly
A higher-usage Business seat for teams with heavier agentic workloads.
- Five times more usage than Standard as currently advertised
- No five-hour limit
- Optional flexible credits
- Same need to confirm preview access and security usage
ChatGPT Enterprise
Custom pricing
For organizations needing advanced governance, contractual controls and flexible usage.
- Codex Security research-preview eligibility
- RBAC and SCIM-synced group controls
- Custom retention and enterprise data protections
- Credit-based and token-based options may apply
ChatGPT Edu
Custom pricing
Institutional access for universities with administrative controls.
- Codex Security research-preview eligibility
- Workspace permissions and RBAC
- Codex Cloud and Codex Security must both be enabled
- Contact OpenAI for institutional terms
Pricing checked . Check current pricing at the source ↗
Assessment
Codex Security strengths and limitations
Where it stands out
- Builds a repository-specific threat model instead of treating every code pattern as equally important
- Makes threat-model assumptions visible and editable by the security team
- Attempts to reproduce potential vulnerabilities before surfacing validated findings
- Provides attack-path and execution evidence that can improve triage quality
- Proposes root-cause patches without automatically modifying the codebase
- Fits findings and fixes into a human pull-request workflow
- Can revalidate a merged remediation to close the loop
- Enterprise and Edu controls can separate use from scan administration
What to consider
- Codex Security is explicitly a research preview and its behavior, availability and pricing can change
- It currently connects directly to GitHub repositories, which may exclude teams on other source-control platforms
- The initial scan of a large project can take longer than subsequent scans
- A flawed or incomplete threat model can direct attention away from the most important risks
- Sandbox reproduction does not prove a finding is exploitable in every production environment
- Failure to reproduce a vulnerability does not prove the code is safe
- Language-model reasoning can still produce false positives, false negatives, incorrect severity and incomplete attack paths
- Generated patches can introduce regressions, weaken other controls or misunderstand intended business behavior
- The product does not use fuzzing or signature-based scanning, so it should complement rather than replace those methods
- It is not a substitute for software-composition analysis, secret scanning, cloud configuration review, dynamic testing or penetration testing
- Repository access and history may expose proprietary code, secrets or sensitive architectural context if permissions are misconfigured
- Pro users should review ChatGPT data controls; Business, Enterprise and Edu data are not used for training by default according to OpenAI
- Finding disclosure, proof-of-concept handling and vulnerability sharing need an internal policy to avoid creating new risk
- Security ownership remains with the organization, including triage, remediation deadlines, exceptions and release decisions
- No separate public price or per-repository allowance makes cost forecasting difficult without testing the eligible workspace
- A clean scan is not evidence that the application has no vulnerabilities
Compare
Codex Security alternatives
The right alternative depends on the specific output, workflow, controls and budget your project requires.
Coding
Claude Security
Choose Claude Security to compare another agentic vulnerability-scanning and patch workflow, including its supported integrations and beta boundaries.
Explore Claude Security →Coding
DryRun Security
Choose Dryrun Security for pull-request-focused application security and code-context analysis within supported repository workflows.
Explore DryRun Security →Coding
Claude Code Review
Choose Claude Code Review when the broader need is automated pull-request review and code-quality feedback rather than dedicated vulnerability discovery and validation.
Explore Claude Code Review →Questions
Codex Security FAQs
What is Codex Security?
Codex Security is OpenAI's research-preview application-security agent. It builds a threat model, scans a GitHub repository and its history, investigates attack paths, validates suspected vulnerabilities and proposes patches for human review.
Who can access Codex Security?
OpenAI's current help article lists ChatGPT Pro, Business, Enterprise and Edu users, subject to rollout, workspace permissions and applicable usage limits.
Does Codex Security automatically change code?
No. It proposes a patch that a human can review and turn into a pull request. The normal review, testing and merge process still applies.
How does vulnerability validation work?
Codex Security attempts to reproduce a suspected issue in an isolated environment and records execution and proof-of-concept details. Reviewers still need to judge whether the assumptions match production.
Does Codex Security replace SAST, DAST or penetration testing?
No. OpenAI says it uses model reasoning and tools rather than fuzzing or signature-based scanning. A mature program should combine it with complementary automated controls and qualified human testing.
How much does Codex Security cost?
There is no separate public Codex Security price. Access is tied to eligible ChatGPT subscriptions, and actual availability, limits or additional credits should be confirmed in the account or enterprise agreement.
Can teams edit the generated threat model?
Yes. OpenAI says teams can inspect and edit it so entry points, trust boundaries, assets and deployment assumptions match the real system.
How should a team pilot Codex Security?
Start with a small repository set and dedicated reviewers. Measure confirmed findings, false positives, missed benchmark issues, triage time, patch quality, regressions, access-control fit and credit consumption before wider rollout.
Bottom line
Our Codex Security verdict
Codex Security is a promising addition for teams that want deeper context and validation than a conventional alert feed provides. Editable threat models, attack-path evidence, sandbox reproduction and reviewable patches create a sensible closed loop. Its value depends on disciplined human review and a layered security program: use it to add coverage and reduce triage effort, never to certify that a repository is safe.
Visit Codex Security website ↗Similar Agents Tools
View all Agents AI tools →
Luma Agents
Luma Agents
My Computer
My Computer - Manus' new desktop app bringing its AI agent to your local machine
Hermes Agent
Hermes Agent - Nous Research's open-source autonomous agent with persistent memory and cross-platform messaging
NemoClaw
NemoClaw - Nvidia's open-source security and privacy layer for OpenClaw autonomous agents
AI training for the future of work.
Get access to all our AI courses, hundreds of real-world AI use cases, live expert-led workshops, an exclusive network of AI early adopters, and more.
AI Courses
Get unlimited access to all of our current & upcoming industry-specific AI courses for the duration of your subscription.
Daily Guides
To keep up with the rapid pace of AI, our team publishes AI implementation guides daily. Our library contains 300+ practical use cases to automate real-world work.
Workshops
Join weekly, live, interactive sessions with industry leaders who are at the forefront of AI for hands-on implementation guidance and exclusive insights.
Community
Network with an exclusive community of AI-first professionals who are working smarter with AI. Learn how early adopters are using AI in their work and businesses.