Aardvark
Free
Aardvark is OpenAI's autonomous agent for code warehouse security. It is oriented to
Aardvark
Core parameters and statistics
Aardvark is an intelligent security protection research assistant launched by OpenAI. Its core task is not to write ordinary application code, but to continuously analyze the code warehouse, identify vulnerabilities, assess exploitability, sort severity levels, and output patching plans. It strings security research, verification and repair into an automated link, focusing on serving teams that already have mature code bases and security review processes.
| Projects | Public Information |
|---|---|
| Official Positioning | Intelligent Security Protection Research Assistant |
| Technical Base | GPT-5 |
| How to work | Continuously analyze the warehouse, submit for scanning, verification, and repair |
| Integration objects | GitHub, Codex, existing engineering processes |
| Public status | Under beta |
| Published price | Undisclosed |
| Free and open | Free scanning is planned for selected non-commercial open source projects |
| Public results | 92% of known and human-injected vulnerabilities identified in benchmarks |
| Announcement time | 2025-10-30 |
Positioning Meaning: The core value of Aardvark is to move the "discovery of vulnerabilities" to the code evolution stage, rather than waiting for penetration testing or post-auditing. For security teams, this means finding problems earlier, completing verification faster, and making it easier to inject fixes back into the development process.
Product Boundary: It is not a daily coding assistant for individual developers, nor is it a general Q&A tool; it is best suited for organizations with warehouse scale, permissions governance, and security response processes.
Workflow Features: The official clearly states that it does not rely on fuzz testing or traditional software component analysis, but uses LLM-driven reasoning and tool calls to understand code behavior, which determines that it is more like a "security researcher collaboration" than a static scanner.
User and market recognition
Internal verification: OpenAI publicly stated that Aardvark has been running in the internal code base for several months and has discovered security vulnerabilities with actual impact, indicating that it is not just a concept demonstration, but has entered real workflows.
External Cooperation: Officials also mentioned that it operates with external alpha partners, and the partners emphasize its depth of analysis and are particularly good at discovering problems that occur under complex conditions.
Open Source Contribution: In open source projects, Aardvark has helped discover and responsibly disclose multiple vulnerabilities, 10 of which received CVE identification, which is a strong signal of recognition by the market and security community.
Quantitative results: OpenAI said it identified 92% of known and human-injected vulnerabilities in the "gold" code base benchmark, which shows that it has a high recall rate for this niche task of security research.
Cost advantage
Aardvark’s public cost structure is very restrained: the official does not show a unified SaaS price list, but gives clear usage boundaries. For enterprises, the true cost is often not just the subscription fee, but also access to the warehouse, manual review, repair problems and security process management.
| Hierarchy | Disclosure | Cost Implications |
|---|---|---|
| C-side/Personal | Undisclosed | Currently not a standard tool open to individuals |
| Developer / API | Not public | The public page does not display the API billing terms |
| Enterprise / Partner | Open for internal testing, application required | Prices and terms are subject to business communication |
| Open source projects | Selected non-commercial projects can be scanned for free | Suitable for open source maintenance and supply chain protection |
Cost Judgment: Its cost advantage is not in low price, but in placing high-value security research capabilities into the development cycle. For large teams, this type of front-end capability can reduce the rework of bug fixes and the cost of downstream incidents.
Main functions
- Continuous Warehouse Analysis: Automatically read and analyze the source code warehouse, and continuously track code changes.
- Vulnerability Assessment: After identifying vulnerabilities, further determine the exploitability and severity level to reduce priority differences among security teams.
- Commit-level scanning: Checks changes as new commits arrive, helping to discover incremental vulnerabilities.
- Sandbox Verification: Try to trigger potential vulnerabilities in an isolated environment to confirm whether they are truly exploitable.
- Patch Generation: Submit repair suggestions and patches generated by Codex to manual review.
- Engineering Collaboration: Works with GitHub, Codex, and existing development processes without requiring teams to change the entire way they work.
Model and version evolution
Current main line
- Aardvark Closed Beta: The official release on 2025-10-30 marks its move from internal and alpha to wider closed beta.
- GPT-5 driver: The release page clearly positions it as a security research assistant that provides technical support for GPT-5, indicating that its capability baseline comes from the capabilities of the current main models.
Public Milestones
- Internal and alpha stage: Officials say it has been running in OpenAI’s internal code base and external alpha partners for several months, indicating that preliminary verification has been completed before the official internal test.
- Open source project expansion: The public page also mentions plans to provide free security scanning for selected non-commercial open source projects, indicating that the product route is expanding from internal pilots to community-side verification.
Technical advantages
Clear mechanism link: Aardvark adopts a multi-stage process of "Analysis -> Submit Scan -> Verification -> Repair". The effect is that the vulnerability does not just stay in a static alarm, but becomes a verifiable and repairable closure.
LLM + Tool Call: It understands behavior by reading code, writing tests, running tests and calling tools, and can cover logical defects, complex conditional vulnerabilities and privacy issues that are difficult for traditional scanners to handle.
Explainable fixes: Each detection result comes with a Codex-generated patch scanned by Aardvark and supports manual review, making fix suggestions easier to be accepted by the development team.
Reduced False Positives: Officials emphasize that the verification phase confirms exploitability in an isolation sandbox. This design can reduce the risk of security teams being overwhelmed by false positives.
How to use
| Entrance | Method | Fit for Task |
|---|---|---|
| Beta application | Submit internal beta application | Obtain partner trial qualification |
| GitHub integration | Access to repositories and commit streams | Continuous scanning for code changes |
| Codex collaboration | Put fixes into the review chain | Accelerate fixes |
| Open Source Project Plan | Request a Non-Commercial Open Source Scan | Supply Chain and Community Security |
Standard process: First apply for internal testing, then connect the target warehouse to the scanning link, then let Aardvark conduct analysis and verification, and finally the engineer will review the patch and merge it into the code.
Key Points for Acceptance: The first pilot should focus on the actual vulnerability discovery rate, false positive control, patch integration rate, and team review burden, rather than just scanning speed.
Product Pricing
Public pricing for Aardvark has not been disclosed, and it is currently closer to a controlled private beta and partner trial model.
- C-side/Personal: Undisclosed, currently not a standard personal subscription product.
- Enterprise/Partner: Enter private beta by applying, price and contract terms are not disclosed.
- Open Source Projects: Official clear plans to provide free security scanning for selected non-commercial open source projects.
Application scenarios
- Enterprise Code Base Security: Suitable for moving security checks forward to daily development and merge stages.
- Open Source Maintenance: Suitable for assisting project maintainers to discover vulnerabilities and make responsible disclosures.
- Complex defect location: Suitable for finding logical vulnerabilities or privacy issues that will only be triggered under specific conditions.
- Fix recommendation generation: Suitable for patches and vulnerability reports to enter the manual review process together.
Applicable people
- Security Engineering Team: Requires continuous scanning, verification, and remediation of closed organizations.
- Platform Engineering Team: The team that needs to integrate security governance into GitHub and engineering processes.
- Open Source Maintainer: Project maintainers who need to discover and disclose vulnerabilities faster.
Unsuitable boundaries are: teams without a formal code repository, no review process, or teams that only need a one-time static scan. This type of requirement is lighter and does not require the introduction of a security collaboration body such as Aardvark.
Summary and Outlook
Aardvark's positioning is very clear: it advances AI from "helping you write code" to "helping you keep code." In the actual development environment, the value of this kind of pre-emptive security capability is usually higher than that of a single scan, because it can embed vulnerability discovery, verification and repair into the engineering process.
Restrictions that still need attention include the scope of open beta, public pricing, enterprise terms and boundaries with existing security tools. For teams preparing for a pilot, it is recommended to start with a single repository or open source project to verify the false positive rate, patch quality and review cost before deciding whether to expand to more code bases.
Related tools: github-copilot, cursor
Version Info
- Aardvark beta version ：OpenAI officially released Aardvark on 2025-10-30 and positioned it as an intelligent security protection research assistant. It is still in the internal testing stage.
- Internal and partner alpha phase ：Official description Aardvark has been running in OpenAI's internal code base and external alpha partner environments for several months to validate vulnerability discovery, manual review and remediation processes.
User Reviews