OpenAI Launches Codex Security to Find, Patch Code Vulnerabilities

OpenAI has unveiled a new AI agent designed to find security flaws and vulnerabilities in enterprise IT systems, positioning it as a competitor to Anthropic’s Claude Code Security launched last month.

Codex Security can scan a company’s IT system, identify flaws, provide a list of solutions, and then fix the vulnerabilities. OpenAI said the agent will aim to provide easy-to-apply patches, with an emphasis on avoiding creating more work than a cybersecurity firm would normally require.

The agent builds on OpenAI’s Aardvark, which was released last year to a small number of beta testers. From the initial test to the research preview, OpenAI says it has reduced noise, the rate of findings, and false positives by over 50 percent, according to the company’s press release.

It has also improved communication between the user and chatbot, allowing it to onboard systems more quickly and analyse code.

The new agent will most likely run on OpenAI’s latest model, GPT-5.4, designed to handle professional work more efficiently than earlier models.

As Codex Security is embedded into more enterprise IT systems, it should continue to improve in both finding vulnerabilities and providing best-practice solutions.

The agent will be rolled out as a research preview for ChatGPT Pro, Enterprise, Business, and Edu customers in the coming days. It is not clear if this will come to free users, as OpenAI is already considering throttling or ending Codex service for them.

The Anthropic vs OpenAI battle intensifies

While Codex Security was always expected to arrive, the fact that it comes a month after the launch of Anthropic’s own version highlights the growing intensity between the two AI giants. Both sides are launching new agents, models, and apps almost weekly, with the other tending to release a similar version a month or two later.

Anthropic has been snatching a few wins from OpenAI in the enterprise market over the past few months. Claude Opus 4.5 was hailed as the new best-in-class model for coding and complex technical work, with hundreds of integrations into popular enterprise software. It has also seemingly taken the mantle from OpenAI of moving public markets, with the launch of its financial agent leading to a drop in several popular fintech service stocks.

Even with this increase in competition, OpenAI is still reportedly ahead in the enterprise market. The last time it published an update on total business customers, it said it had over 1 million, while last year, Anthropic had around 300,000. The gap has narrowed since then, however, and that is without even considering the growth of Google Gemini in the market.

If the enterprise market is heating up, OpenAI appears safer on the consumer side. It is nearing one billion active users on ChatGPT, far more than Claude. Google has said more than 650 million people are using Gemini, although it is heavily promoting the service on its search engine and web browser, which both have over three billion users.

OpenAI recently closed the largest funding round ever, raising $110 billion at a valuation of $730 billion, while Anthropic was valued at just over half of that, $380 billion, in February this year.

The post OpenAI Launches Codex Security to Find, Patch Code Vulnerabilities appeared first on eWEEK.