1300 166 138 Tell us what you need
News / Security

Why ‘vibe coding’ is becoming a security problem for business

By Monika Tantau • Published 5 months ago
Why 'vibe coding' is becoming a security problem for business

Key takeaways

• An end-of-financial-year briefing warns AI is accelerating both cyberattacks and insecure code
• A cited test found leading generative AI models failed to produce consistently secure code
• The advice is to move from periodic scans to continuous, automated exposure monitoring
• Australian mid-market teams are urged to scan AI-written code inside their CI/CD pipelines

Most Read

As Australian organisations close the books on the financial year, a security briefing is urging leaders to treat 1 July as a reset for how they defend their systems. Writing for Technology Decisions, Zak Menegazzi, APJ director at Armis (a ServiceNow company), argues that AI has changed both sides of the security equation at once.

Attackers are using it to find and chain small weaknesses at speed, while internal teams are shipping AI-generated code that carries flaws straight into production. The result is a widening gap between how fast vulnerabilities appear and how fast most businesses can close them.

What the briefing actually found

The most pointed claim concerns code quality. The briefing cites testing, the source of which was not disclosed, that found prominent generative AI models failed every time to produce consistently secure code. Menegazzi describes the pattern as “vibe coding”: developers accept AI output that looks right and runs on the surface, while missing resource limits, memory buffer overflows and similar flaws slip into core enterprise systems during rapid deployment cycles.

This is not a hypothetical risk. As more teams lean on AI assistants to ship faster, the volume of code entering production rises, and so does the share of it that no human has read line by line.

Why periodic patching falls behind

Traditional security programs were built around scheduled scans and reactive patching. The briefing argues that cadence no longer matches the threat. When a vulnerability can be discovered and weaponised within minutes, a monthly or quarterly scan leaves a window wide enough to drive an attack through.

AI compounds the problem by helping adversaries link minor, individually low-risk weaknesses into a viable path to critical systems. A flaw a scanner might rank as trivial becomes serious once it is one step in a chain.

The case for autonomous defence

The recommended shift is from periodic scanning to continuous exposure monitoring, backed by automation that runs at machine speed. In practice that means a real-time inventory of every asset, prioritisation based on what is genuinely exploitable and business-critical rather than raw severity scores, and a detection-to-remediation loop that does not wait for someone to file a ticket.

It also means moving security earlier into the development pipeline. Integrating AI-aware checks into CI/CD lets teams scan AI-generated code with the same rigour they would apply to anything else before it reaches production.

The credential basics that still decide breaches

Faster, AI-assisted attacks make the unglamorous fundamentals more important, not less. Many breaches still begin with a weak, reused or phished password, and an automated attacker will test those at scale long before it attempts a clever code exploit.

For most mid-sized teams, the highest-return controls remain a company-wide password manager and enforced multi-factor authentication. We have covered how to choose one in our guide to password managers for Australian businesses, and the trade-offs between the leading options in our Bitwarden and 1Password comparison. Autonomous detection matters, but it works best on top of credential hygiene that closes the easiest door first.

The credential basics that still decide breaches

What this means for Australian businesses

For Australian mid-market firms, the new financial year is a practical moment to review where AI now sits in both the attack surface and the toolchain. The Australian Cyber Security Centre’s Essential Eight already pushes organisations toward patching, application control and multi-factor authentication; the briefing’s argument is that the speed of AI-era threats raises the bar on how continuously those controls need to run.

There is also a governance layer. Under the Privacy Act, organisations remain accountable for data exposed through insecure systems, including code written by an AI assistant. That accountability does not transfer to the tool. The defensive side is being rebuilt around AI too, not just the offensive one, as we noted when Australian cyber teams gained access to Anthropic’s Mythos model earlier this month.

The discipline that defines the year ahead

The throughline is speed. AI has compressed the time between a weakness appearing and an attacker using it, and the briefing’s case is that defence has to be compressed to match. For businesses adopting AI across their operations, building security review into how that AI is used, rather than bolting it on afterwards, is shaping up to be the cyber discipline that separates the prepared from the exposed this year.

Scroll to Top