AI Is Reshaping Cybercrime: What Perth Businesses Need to Know

A report published by security vendor ThreatDown (Malwarebytes) in July 2026 makes a blunt argument: cybercrime is being rebuilt around AI, faster than most organisations realise. Most of the specific incidents it documents happened at nation-state and enterprise scale, not to businesses our size. But the shift underneath those incidents, cheaper attack tooling and unmanaged AI adoption, is relevant to any Perth business, and worth unpacking without the hype.

What the Report Actually Found

Stripped of the more dramatic framing, the report's research points to a few concrete, attributable findings:

• Malicious AI is sold openly, not hidden. The report's researchers mapped a marketplace of AI tools built for cybercrime, indexed by Google, with pricing pages and checkouts like any legitimate SaaS product. Most of these tools turned out to be jailbroken wrappers around mainstream AI models rather than anything custom-built.
• Guardrail-stripped open-source models are spreading fast. The report found thousands of AI models published openly under labels like "uncensored" or "abliterated", models modified to answer requests a mainstream AI would refuse, with tens of millions of downloads in a single month. Run locally, there's nothing for a security tool to monitor or block.
• Malicious AI "skills" are a new attack category. As AI agents that can install plugin-style extensions have grown in popularity, the report documents extensions built to steal credentials directly, and others built as social-engineering lures that trick the person running the agent instead of the agent itself.
• Shadow AI is already showing up in breach costs. The report cites Netskope's 2026 Cloud and Threat Report and IBM's Cost of a Data Breach 2025 findings that close to half of employees using generative AI at work do so through personal, unmanaged accounts, and that breaches linked to shadow AI cost organisations more on average than a standard incident.
• AI is getting good at finding software flaws. Google's Threat Intelligence Group reported identifying what it believes is the first zero-day exploit built with AI assistance by criminals. Separately, Anthropic has reported testing an AI model capable of finding vulnerabilities in every major operating system it tried, and chose not to release it publicly because of that capability.

Why This Matters Even If You're Not a Target of Nation-State Hackers

The specific breaches named in the report, attacks on government agencies and critical infrastructure, aren't the direct risk to a Perth SMB. What matters is what those incidents reveal about the tools now available further down the market:

• Phishing and social engineering keep getting harder to spot. The same AI-assisted attack tooling covered in this report is what's behind the trend we've written about before in how AI is changing cybersecurity for Perth businesses - better-written phishing, cloned voices, and faster reconnaissance, now available to anyone willing to pay a subscription fee for it.
• Patch backlogs get more expensive to carry. The report notes Firefox shipped 423 bug fixes in April 2026, compared with 31 in April the year before, after applying an early AI vulnerability-finding tool to its own codebase. If AI-assisted tools keep surfacing vulnerabilities faster than they used to, a business without a consistent patch management routine falls further behind, not less.
• Shadow AI is a live identity risk, not a theoretical one. Every AI account an employee creates without IT's knowledge is another set of credentials, and potentially another connection to your files and systems, that nobody is managing. This is the same gap we cover in setting an AI usage policy for staff.

Three Things Worth Doing Now

The report's own recommendations translate reasonably well down to SMB scale:

• Patch on a schedule, not "when we get to it". Automated patching and a regular vulnerability review close the gap that unpatched systems leave open the longest. An IT security audit is a practical way to find out where that schedule is currently slipping.
• Make sure something is actually watching, not just installed. Antivirus alone won't catch AI-assisted attacks built to avoid known patterns. Modern endpoint detection and response, paired with someone who actually reviews the alerts, is what catches the early signs of a breach before it spreads.
• Find out what AI tools your staff are actually using. You can't govern what you can't see. Start with a written AI usage policy, then check your own AI governance more broadly with our free 2-minute AI readiness self-assessment.

Frequently Asked Questions

For related reading, see How AI Is Changing Cybersecurity for Perth Businesses, What Is Agentic AI Security, and Patch Management for Perth Businesses.

This article summarises third-party research and reporting published by ThreatDown (Malwarebytes) in its "Cybercrime in the Age of AI" report, along with sources that report itself cites (including Google's Threat Intelligence Group, Anthropic, Netskope, and IBM). It is general information only, not a formal security assessment, and reflects reporting available at the time of writing. Figures and claims about specific incidents, vendors, or products are those of the original researchers, not independently verified by Top Tier Computing.