AI-written code blamed for 1 in 5 major breaches, new report shows

AI code linked to major breaches

Aikido Security’s State of AI in Security and Development survey found that about one in five organizations reported a major security incident they linked to AI-generated code.

Companies are eager to use AI to speed up development, but vulnerabilities in AI-generated code are proving widespread and costly.

Aikido’s survey respondents reported that roughly 24 percent of production code is now AI-generated, which is complicating accountability for defects and breaches. Security, development, and legal teams are unsure who is responsible when AI code causes problems.

AI-generated code brings big risks

Research finds that 69 percent of organizations discovered vulnerabilities in AI-generated code. About 20 percent of respondents told the survey they had experienced a serious incident they associated with AI-generated code.

Blame is shared among security teams, developers, and those merging code. This confusion makes managing AI-induced risks more difficult for organizations globally.

Who’s accountable for AI mistakes?

AI-generated code creates a real accountability dilemma. Security, development, and merger teams each risk being blamed when breaches occur, creating uncertainty in organizations. Mike Wilkes from Aikido called it a “real nightmare of risk.”

The survey found that US respondents reported more serious incidents than European respondents, and the authors suggested this may reflect differences in compliance regimes and developer practices, such as bypassing security controls.

Humans still vital for AI

Only 21 percent of organizations believe AI can operate without human oversight. Security and development teams remain crucial for reviewing AI-generated code and ensuring compliance.

AI is a tool to enhance productivity, not a replacement for humans. Proper oversight helps organizations manage vulnerabilities and maintain accountability.

Security teams under pressure

When AI-generated code causes issues, 53 percent of respondents say security teams get blamed. Developers are also at risk, with 45 percent facing scrutiny for errors.

The shift highlights challenges in shared responsibility. Companies need clear protocols to prevent friction and protect both human and system performance.

Developers feel the heat

Half of developers believe they’d be blamed if AI-generated code introduced a vulnerability, often more than the security team. The added pressure complicates development workflows.

Developers must continuously monitor AI outputs and coordinate with security to ensure automated code does not introduce new risks or breaches.

AI optimism meets reality

While 96 percent of organizations believe AI will eventually produce secure code, the timeline averages over five years. Optimism is tempered by current vulnerabilities and oversight needs.

AI is also expected to handle penetration testing in roughly 5.5 years, but nearly all respondents agree humans will continue to play a critical role.

AI can help but has limits

AI tools improve efficiency but are not flawless. Organizations report vulnerabilities, showing human supervision remains essential to catch mistakes before they escalate.

Recognizing AI’s limitations helps organizations adopt safer workflows. Combining AI productivity with human expertise maximizes security and reliability.

Ownership confusion creates delays

Organizations often struggle to assign responsibility for AI errors. Confusion over whether security, development, or legal teams are accountable delays remediation and heightens risks.

Clear ownership policies ensure faster detection and resolution of AI-generated vulnerabilities, reducing operational risk and improving accountability.

US vs Europe breach trends

AI-related breaches are more common in the US than Europe, partly due to less strict compliance and developers bypassing security controls. European companies report fewer serious incidents but still face near misses.

Understanding regional differences helps companies tailor AI governance and training. Local policies and habits influence how safely AI code is deployed in production.

Preparing for AI-driven security

Organizations need clear policies, streamlined AI tools, and strong human oversight to mitigate risks. Preparing for AI-driven workflows now prevents future breaches and confusion.

Aligning teams, defining ownership, and monitoring AI code allows companies to harness AI productivity without compromising security. The next wave of software development depends on this balance.

AI breaches highlight urgent lessons

The rise of AI-caused breaches shows that organizations cannot blindly trust automated code. Human review, accountability, and risk management remain vital for secure systems.

Learning from current incidents prepares companies for safer AI adoption. Proactive steps help prevent AI-induced breaches from causing serious damage in the future.

Are AI tools helping or hurting tech jobs? See how Microsoft layoffs hit programmers as AI writes more code.

AI security future looks balanced

Despite current challenges, optimism remains. Nearly all organizations expect AI to write secure code and handle penetration testing within five years, with humans continuing oversight.

This balance between AI efficiency and human judgment promises a safer, productive future. Companies can leverage AI without sacrificing control or accountability.

Is AI helping developers or taking over coding jobs? See how Microsoft’s AI wrote nearly a third of its code.

Do you think AI-written code is a security risk, or are breaches inevitable? Share your thoughts and drop a like if you found this important.

Read More From This Brand:

• a16z finds startups spending on vibe coding
• How vibe coding helped a Swedish AI startup break every growth record
• Google AI Can Now Build Apps From Your Words?

Don’t forget to follow us for more exclusive content right here on MSN.

If you liked this story, you’ll LOVE our FREE emails. Join today and be the first to get stories like this one.

This slideshow was made with AI assistance and human editing.