AI Can Find Vulnerabilities. Who Will Fix Them?

OpenAI's June 22 Daybreak expansion and Codex Security updates.AI is making software development faster, but it is also changing the way we approach software security.For years, security engineering has involved a combination of automated scanning, manual code review, penetration testing and continuous monitoring.These tools are useful, but finding a vulnerability is only one part of the problem.A security team still needs to determine whether the issue is real, understand its impact, develop a fix, test that fix and make sure it does not introduce another problem.
AI could potentially help across this entire process.
In June, OpenAI expanded Daybreak, an initiative focused on helping security teams discover and patch vulnerabilities. The announcement included updates to Codex Security and the broader availability of GPT 5.5 Cyber for trusted defenders.
From Detecting Problems to Fixing Them
Traditional security scanners generally identify patterns that might indicate a vulnerability.For example, they might flag an unsafe dependency, an exposed endpoint or an insecure way of handling user input.But identifying a suspicious pattern does not always establish whether the application is actually vulnerable.
The surrounding code matters.A particular function might look unsafe in isolation but be protected by another layer of validation. Alternatively, a vulnerability might only become exploitable when several components interact.This is where AI assisted security analysis becomes interesting.
A model can inspect a larger section of the codebase, understand relationships between components and investigate possible attack paths.It can then help generate a patch that developers can review.
Why Automated Patching Is Difficult
Generating a patch is not the same as fixing a security issue.A patch needs to address the actual vulnerability without breaking existing functionality.Consider an authentication system. A security fix might involve changing how tokens are validated or how permissions are checked.A careless change could prevent legitimate users from accessing the application or introduce another weakness.This is why validation matters.Generated patches should be tested against the relevant application behaviour, security requirements and existing test suites.Human review also remains important, particularly for changes involving authentication, authorisation and sensitive data.
Security Is Becoming Part of the Development Workflow
One of the more interesting implications is that security tools could become more closely integrated into everyday development.Instead of waiting for a separate security review, developers could receive useful feedback while writing code or preparing a pull request.AI systems could help explain why a particular implementation is risky and suggest a safer alternative.This could reduce the distance between identifying a vulnerability and fixing it.However, it also means teams need better processes for reviewing AI generated security changes.
The objective should not be to automatically approve every suggested patch.
The Bigger Question
I think the movement from vulnerability detection towards vulnerability remediation is an important development.As AI helps developers produce code faster, security engineering needs to evolve alongside it.The goal should not be to generate more alerts or create more automated reports.It should be to help teams understand risks earlier, fix them correctly and ship fewer vulnerabilities into production.