OpenAI open-sources its vulnerability-hunting tool, and hands it to anyone with a terminal
Codex Security CLI is Apache 2.0 licensed, scans repositories, verifies fixes and plugs into CI pipelines. It is the same system that helped patch 3,000 critical vulnerabilities.
OpenAI has released Codex Security CLI as open source under the Apache 2.0 licence. It is a command-line tool that helps security and development teams find, confirm and fix vulnerabilities in code repositories. It can scan a repository, compare results across multiple runs, verify that a fix actually worked, and hook security checks into CI/CD pipelines, the automated systems that test and ship code every time a developer commits a change. Bulk scans across multiple repositories are supported. The tool needs Node.js 22 and Python 3.10 or higher, installs via npm, and is currently in beta.
This is not a new system with a new name. Codex Security was known internally as “Aardvark” and launched in March 2026 as a research preview limited to ChatGPT Enterprise, Business and Edu customers. By April, OpenAI said it had helped fix more than 3,000 critical vulnerabilities. Moving it to an open licence takes it out of the enterprise tier and puts it in reach of solo maintainers, small teams and the open-source projects that most of the internet quietly depends on.
The competitive picture is straightforward: Anthropic ships Claude Security, which also scans codebases and suggests patches. Both companies are building the same thing for the same reason, and that reason showed up in the news twice this month. AI models are getting good at finding software vulnerabilities, which is useful when defenders do it and alarming when attackers do. OpenAI’s own models breached Hugging Face during an internal evaluation by exploiting a zero-day in a cache proxy. If automated attack capability is now real, automated defence has to be at least as widely available, or the asymmetry gets ugly fast.
Worth naming the caveat plainly: a tool that finds vulnerabilities also teaches you where vulnerabilities are. Open-sourcing it lowers the bar for everyone, defenders and attackers alike. The bet OpenAI and Anthropic are both making is that defenders benefit more from broad distribution than attackers do, because attackers with real resources already have these capabilities and defenders mostly do not. That bet is reasonable. It is still a bet.
What this means for you: If you do not write code, nothing here touches your day, but it is a useful data point for a common worry. The question “will AI make hacking easier” now has a visible counterweight: the same labs are shipping the defensive version for free. If you do write code, even a small side project with dependencies, this is worth an evening. It is in beta, so expect rough edges and false positives, and treat its findings as leads to investigate rather than verdicts. Run it once against something you maintain and see what it says.
Sources
OpenAI's new transcription models are faster and 25 percent cheaper, but still not the most accurate
GPT Transcribe and GPT Live Transcribe cut the price to $0.0045 per minute and hit a 3.31 percent word error rate. ElevenLabs, Google and Mistral all score better.
