After Claude, ChatGPT begins to hunt security flaws.

Generative AI is now searching for vulnerabilities. It was Anthropic who first unveiled Claude Code Security, a feature of Claude Code that identifies security flaws in computer code. Thanks to the Opus 4.6 model, AI identified over 500 critical zero-day vulnerabilities in open source projects, including 22 in Mozilla's Firefox browser.

We partnered with Mozilla to test Claude's ability to find security vulnerabilities in Firefox.

Opus 4.6 found 22 vulnerabilities in just two weeks. Of these, 14 were high-severity, representing a fifth of all high-severity bugs Mozilla remediated in 2025. pic.twitter.com/It1uq5ATn9

— Anthropic (@AnthropicAI) March 6, 2026

Among these vulnerabilities, 14 are considered critical, which is a fifth of all critical vulnerabilities fixed by Mozilla in 2025, notes Anthropic. This is a true revolution in the field of security flaw detection.

Also read: Why everyone uninstalls ChatGPT and downloads Claude

Codex Security, OpenAI's security agent

Determined to rival Anthropic on all fronts, OpenAI quickly announced a similar function, Codex Security. The tool is described as a security agent capable of automatically scanning large code repositories to uncover vulnerabilities and bugs. The « application security » agent analyzes « in-depth your project to identify complex vulnerabilities that other similar tools do not detect, and proposes reliable solutions that significantly improve the security of your system while avoiding disturbing you with minor bugs », explains OpenAI in its announcement.

Codex Security—our application security agent—is now in research preview.https://t.co/JG2uwGUJFv

— OpenAI (@OpenAI) March 6, 2026

To reduce false positives, Codex Security develops real exploitation techniques. These exploitation methods, deployed in an isolated environment, should allow verifying that a vulnerability is indeed present in the code. If the vulnerability is confirmed, the agent proposes a fix tailored to the code's context. This fix is designed to avoid breaking existing working code and to remain consistent with the entire system architecture. Over the past 30 days of testing, OpenAI's agent has scrutinized over 1.2 million code changes from open-source projects. It has identified 792 critical vulnerabilities and over 10,000 serious ones.

Also read: ChatGPT and the military – after a controversial agreement, an OpenAI leader resigns by « principle »

The question of vulnerability exploitation

On paper, Codex Security does better than its rival in the field of creating exploitation techniques. While OpenAI's tool uses AI to generate functional proof-of-concepts directly within the context of the running system, Claude Code still struggles to find a way to exploit vulnerabilities. According to tests conducted by Anthropic, AI generally fails to find exploits. After testing it hundreds of times, AI has only managed to produce a real working attack program for two vulnerabilities.

In any case, Codex Security and Claude Code Security represent serious competitors to security companies. Following the release of both tools, the stock actions of cybersecurity giants have significantly contracted. Just after its release, Claude Code Security caused a temporary stock market crash for key groups like CrowdStrike, which lost 8% in a single session on Wall Street, Fortinet, Okta, Cloudflare, Zscaler, Datadog, and SentinelOne.

Also read: Mass unemployment, layoffs, and AI – a shocking report sows panic on financial markets

A week later, the arrival of Codex Security intensified pressure on these stock values, with several actions turning red again. The American investment bank Baird refers to a panic sell-off triggered by a new market cyber perception.

👉🏻 Follow tech news in real-time: add 01net to your Google sources, and subscribe to our WhatsApp channel.