AI security risks every small business needs to understand

A contractor working with a New South Wales government agency pasted a spreadsheet containing over 12,000 rows of personal data into ChatGPT. No malicious intent. Just a shortcut to get a task done faster. The result was a notifiable data breach affecting thousands of Australians in a flood assistance program. The system that failed wasn’t a firewall or an email filter. It was the absence of a clear policy about what staff can and cannot feed into an AI tool.

AI tools are now part of daily work for a growing number of Australian small businesses. Platforms like Microsoft Copilot, Google Gemini, Anthropic’s Claude, and ChatGPT are being used for drafting, summarising, coding, customer service, and data analysis. That’s largely a good thing. But the speed of adoption has outpaced the security thinking in most small and medium businesses, and the risks are concrete and growing. Australia’s ASD’s ACSC has published guidance specifically for small businesses on AI cyber security risksdeveloped with COSBOA and New Zealand’s NCSC. The risks they highlight are worth understanding before your next AI tool gets approved.

This article covers the five AI security risks that matter most to Australian SMBs right now, and the practical steps that reduce exposure without requiring a large security team. If you want to understand the broader picture of how your current security posture stacks upthat’s a useful place to start.

Why AI has changed the threat picture for small businesses

AI has not created new categories of attack. It has made familiar attacks faster, cheaper, and harder to spot. The ACSC’s Annual Cyber Threat Report 2024-25 documents that AI almost certainly enables malicious actors to execute attacks at a larger scale and a faster rate than was previously possible. The same tools that help a small business write better emails are being used by attackers to write better phishing emails, in multiple languages, at volume, with personalisation that used to take hours of manual research per target.

The cost picture for Australian businesses reflects this shift. The ACSC’s 2024-25 report found the average self-reported cost of a cybercrime incident was around $56,600 for small businesses and $97,000 for medium businesses, with medium business costs jumping 55 percent in a single year. These figures are self-reported, which means actual costs including business disruption, reputational damage, and recovery effort are likely higher. Organised criminal groups target small businesses deliberately, because smaller businesses tend to hold valuable customer data but operate with fewer defences than large enterprises.

AI is sharpening the attacker’s edge on multiple fronts simultaneously. Understanding which risks apply to your business is the first step to doing something about them.

What is shadow AI, and why is it the most common risk for SMBs?

Shadow AI refers to AI tools that staff are using inside a business without IT awareness or approval. An employee pastes a client contract into an online summariser to save time. A developer asks an AI coding assistant for a quick solution and installs a suggested library without checking what it is. A customer service team member uses a free AI chatbot to draft responses containing customer account details. None of these actions are necessarily malicious, but each creates real exposure.

The ACSC identifies potential misuse of customer data by third-party AI providers as a key risk for small businesses using cloud-based AI tools. The problem is structural: when a staff member inputs data into a public AI platform, that data leaves your environment and enters the AI provider’s systems. What happens to it depends on the provider’s data handling and privacy policies, which most people never read. The NSW Reconstruction Authority breach, where a former contractor uploaded personal data into an unauthorised AI tool, is a clear example of how shadow AI creates notifiable data breaches under Australian privacy law without any attacker involvement at all.

The practical fix is an AI use policy that clearly defines what data cannot be uploaded into AI platforms, combined with some basic visibility into which tools staff are actually using. Shadow AI is a management and policy problem as much as a technical one.

How AI-powered phishing is different from what came before

AI-powered phishing is more dangerous than traditional phishing because it removes the signals people have been trained to spot. Generic greetings, obvious spelling errors, awkward phrasing, and implausible premises were the giveaways that phishing awareness training relied on. AI-generated phishing strips those away. It can produce personalised, well-written messages that reference real relationships, real job titles, real recent events, and the correct tone for a given industry.

Attackers are now using AI to automate victim profiling at scale, pulling information from LinkedIn, company websites, supplier lists, and social media to build credible lures for specific individuals. A finance manager at a Brisbane professional services firm might receive a convincing email appearing to come from their firm’s principal, referencing a real client name, asking for an urgent payment. The ACSC reports that business email compromise (BEC) remains one of the most financially damaging attack types for Australian small businesses, and AI makes BEC easier to execute convincingly.

The defence is not better spam filters alone. It requires out-of-band verification for financial requests, multi-factor authentication (MFA) on all email accounts, and regular awareness training that reflects what AI-generated phishing actually looks like. The right email security configuration also reduces the volume reaching staff in the first place.

What is the data leakage risk from AI tools staff are already using?

Data leakage through AI tools happens in two ways. The first is intentional but uninformed: staff putting sensitive data into AI tools to get work done faster, without understanding that the data may be retained, used for model training, or accessible to the AI provider’s staff. The second is accidental: AI tools that are integrated into business systems inheriting permissions and accessing data they were never intended to touch.

Recent research found that around 80 percent of AI-related incidents involve regulated or sensitive data. The issue is not just user behaviour. AI tools inherit access from the systems they connect to, often without restriction. A Microsoft 365 Copilot deployment that has not been properly scoped can surface sensitive documents to users who would not normally have access to them, simply because the underlying permissions were never properly configured. The same applies to other AI integrations that connect to CRMs, HR systems, or financial platforms.

For Australian businesses, data leakage through AI tools creates obligations under the Privacy Act 1988 and the Notifiable Data Breaches (NDB) scheme. The OAIC’s Notifiable Data Breaches scheme requires businesses to notify affected individuals and the OAIC when a data breach is likely to cause serious harm. An AI-related data spill qualifies. The ACSC’s guidance recommends reviewing data handling practices and the privacy policies of AI vendors before deployment, and establishing an internal policy that defines what data cannot enter AI systems.

How does AI-assisted ransomware work differently from older attacks?

AI-assisted ransomware attacks follow the same broad sequence as earlier ransomware campaigns, but each stage is faster and more precise. Attackers gain initial access through phishing, credential stuffing, or a known vulnerability. AI-powered reconnaissance tools then map the network automatically, identifying high-value systems, backup locations, and administrative accounts. Lateral movement happens faster because AI tools help attackers identify the fastest path to the most valuable data. Data is exfiltrated before encryption begins, giving attackers a second lever: pay the ransom, or the stolen data gets published.

The ACSC responded to 138 ransomware incidents in FY2024-25, and ransomware comprised 11 percent of all reported cyber incidents that year. Australia’s mandatory ransomware reporting regime, introduced in May 2025 for businesses with annual turnovers of $3 million or more, means the regulatory and reputational stakes of a ransomware incident have increased. Attackers know this. The regulatory pressure to resolve incidents quickly is now factored into ransom demands and extortion timelines.

Stopping modern ransomware requires controls at multiple stages of the attack lifecycle: securing initial access through strong MFA and patching, detecting unusual lateral movement, segmenting internal networks so one compromised machine doesn’t reach everything, and keeping backups isolated and recoverable. The ACSC’s Essential Eight framework addresses each of these layers and remains the most practical starting point for Australian SMBs.

What is the supply chain risk when using AI-powered vendors?

Supply chain risk from AI tools means that a vulnerability in your AI provider’s systems can become your vulnerability, even if your own systems are secure. Small businesses using AI Software as a Service (SaaS) applications rely on third-party providers to manage the underlying infrastructure, AI models, data handling, and operations. Any weakness in that provider’s environment creates indirect exposure for every business using their platform.

The 2025 Qantas data breach illustrated this principle at scale: attackers compromised a third-party supplier and accessed six million customer records. Qantas’s own systems were not breached. The lesson for SMBs is that your security posture now includes the posture of every vendor you connect to. The ACSC recommends choosing AI and SaaS services that are secure by design and secure by default, and reviewing vendor data handling, liability, and audit rights before connecting any platform to business data.

For businesses in professional services, this risk carries additional weight. Legal, accounting, and medical practices hold data that attracts particularly motivated attackers, and the confidentiality obligations in those sectors mean an AI-related breach carries professional and regulatory consequences beyond the immediate financial cost. If your business operates in one of these sectors, TTA’s IT support for professional services firms addresses the specific controls those environments need.

What are the practical controls that reduce AI security risk for Australian SMBs?

The ACSC’s Essential Eight framework covers the foundational controls that reduce AI-related risk most effectively. Multi-factor authentication, particularly phishing-resistant MFA for admin accounts and remote access, stops most credential-based attacks that AI-powered phishing tries to enable. Application control prevents unauthorised software, including AI-generated code with embedded malware, from running on business systems. Patching removes the vulnerabilities that AI-assisted reconnaissance is designed to find and exploit quickly.

Beyond the Essential Eight, three AI-specific controls make a material difference for SMBs. First, an AI use policy that defines which tools are approved, what data cannot be entered into those tools, and what the process is for reviewing new AI tools before staff start using them. Second, data classification that identifies what information is sensitive or regulated, so that staff and AI tool configurations know which data needs protection. Third, vendor review before connecting any AI tool to business systems, checking data handling policies, retention practices, and what happens to inputs after they are processed.

The honest constraint is that most small businesses don’t have an internal security team to run these controls. The realistic approach for most Australian SMBs is a managed security model that provides monitoring, configuration, and response capability without requiring a full-time hire. That’s the model we run at TTA for clients across South-East Queensland, combining the right tooling with the human oversight that makes it work. Our cyber insurance compliance support helps businesses get the controls in place that insurers and regulators increasingly expect.

Where to start if you’re not sure where you stand

An IT security assessment is the practical starting point for any business that isn’t confident about its current exposure to AI-related risks. It identifies what’s already in place, where the gaps are, and which controls will have the most impact for your specific environment. It also gives you a clear picture of your obligations under the Privacy Act and the Notifiable Data Breaches scheme before an incident forces that conversation.

If you’d like to talk through what AI security risks look like for your business, get in touch with the TTA team. We work with SMBs across Brisbane and South-East Queensland, and we’re straightforward about what’s worth prioritising and what can wait. Start a conversation with us here.