Most businesses are already using AI in some form: a customer-facing chatbot, an AI-assisted productivity tool, or a platform that uses machine learning behind the scenes. The benefits are real. So are the risks.
This article outlines five risk areas worth understanding before and during AI integration in your business. None of these should put you off using AI. But knowing what to look for helps you make smarter decisions and avoid being caught off guard.
Here are five risks to consider when integrating AI tools into your business.
Five AI Risks Every Business Owner Should Know
1. You’re Only as Secure as the Tools You Rely On
When you use a third-party AI solution, you’re not just buying a product. You’re taking on a dependency. If that vendor’s platform goes down, gets hacked, or changes its terms, your business feels it too.
It is sometimes called software supply chain risk, and AI tools have introduced new versions of that risk.
Consider what happened in late 2022, when a popular machine learning library called PyTorch was compromised through a supply chain attack. A malicious package was uploaded to a public code repository under the same name as a legitimate dependency. Before it was caught, it had been downloaded approximately 2,300 to 3,000 times, with each download capable of silently harvesting credentials and sensitive files from the affected machine.
More visibly, ChatGPT (used by millions of businesses globally) experienced multiple significant outages across 2023 and 2024, including a March 2023 incident where a bug in a third-party library briefly exposed fragments of other users’ chat histories, and a December 2024 outage lasting around nine hours following a power failure at a Microsoft data centre.
What this means for your insurance position: Dependence on third-party AI platforms is a risk exposure. When you speak with your insurance adviser, it’s worth discussing how your policy responds to losses caused by a vendor outage or compromise, rather than a direct attack on your own systems.
2. AI Tools Can Become Entry Points for Attack
Any AI tool that sends and receives information (a chatbot on your website, an AI calculator, an automated assistant) creates a channel that can, in some cases, be manipulated.
One technique is known as jailbreaking: prompting an AI model to behave outside its intended parameters. This isn’t always a dramatic hack. Sometimes it’s as simple as a customer asking questions in an unexpected sequence that lead to responses the business did not authorise.
The consequences can range from the embarrassing to the legally significant. In a case decided by the British Columbia Civil Resolution Tribunal in February 2024 (Moffatt v Air Canada [2024 BCCRT 149]), a passenger successfully claimed damages after Air Canada’s website chatbot gave him incorrect information about bereavement fares. The tribunal found Air Canada responsible for the chatbot’s output. The argument that the chatbot was a “separate entity” whose errors Air Canada couldn’t be held liable for was rejected outright.
The practical takeaway isn’t that chatbots are dangerous. It’s that businesses are responsible for the information their AI tools provide. If your AI tool gives a customer incorrect pricing, policy terms, or advice, the fact that a machine said it is unlikely to be a defence.
What this means for your insurance position: Liability for what your AI tools say or do can land with your business, not the vendor. That’s worth understanding when reviewing your professional indemnity and public liability cover.
3. Training AI Requires Data and That Creates Exposure
AI tools get better by learning from data. A lot of data. That means vendors often centralise and aggregate large volumes of information, sometimes including data belonging to their customers, to train and improve their systems.
This concentration of data is a risk in its own right.
In September 2023, Microsoft’s AI research team accidentally exposed approximately 38 terabytes of internal data (including private keys, passwords, and tens of thousands of internal messages) through a misconfigured access token on a public code repository. The misconfiguration had existed undetected for around three years before being discovered by security researchers.
In 2024, cloud data platform Snowflake was at the centre of a major breach affecting around 165 organisations, after attackers used stolen credentials to access customer accounts. The exposed data spanned hundreds of millions of individuals across companies including AT&T, Ticketmaster, and Santander.
Neither incident involved a sophisticated zero-day exploit. Both came down to access controls that weren’t tight enough.
What this means for your insurance position: The concentration of data in AI vendor environments is a risk exposure. It’s worth discussing with your insurance adviser how your cyber policy responds to a breach originating from a third-party vendor, not just your systems.
4. AI-Powered Security Tools Carry Their Own Risks
AI is increasingly used within cybersecurity itself. Tools that monitor networks, detect anomalies, and respond to threats automatically can be genuinely effective. They can also cause problems when they get things wrong.
Because these tools operate at a high level of system access, an error (whether from a misconfiguration, a false positive, or an unexpected interaction with other software) can lock legitimate users out of their own systems, reset credentials without warning, or quarantine infrastructure in ways that make recovery from an actual incident harder, not easier.
On 30 July 2024, Microsoft Azure experienced a significant outage triggered by a Distributed Denial of Service (DDoS) attack. The attack itself was initially contained by Azure’s automated defences, but an error in how those defences disengaged caused traffic to be misrouted, resulting in an outage of approximately eight hours that affected banking, business operations, and Microsoft 365 services globally. The disruption wasn’t caused by the attack getting through. It was caused by the automated response to the attack misfiring.
What this means for your insurance position: AI security tools that operate with high system access create a distinct exposure. An error or misconfiguration in one of these tools could itself trigger a business interruption event. It’s worth reviewing whether your policy covers losses caused by your security systems, not just external attacks.
5. How You Use AI Affects Your Insurance Position
How AI affects your insurance is often an area that surprises many business owners: how you integrate AI into your operations is increasingly relevant to your insurance coverage.
When insurers assess cyber and technology-related risks, they’re looking at the whole picture, including whether the businesses they cover have taken reasonable steps to manage AI-related exposures. They will typically want to understand:
- The protection measures you have in place
- Whether AI tools are being developed, deployed, or tested within the business
- Whether the AI tools or developers centralise data in ways that could increase your risk exposure
- What level of access the AI systems have to your business infrastructure
- How you manage and monitor access, permissions, and user accounts
The answers to these questions can affect whether a policy is offered, on what terms, and at what cost. Introducing a new AI tool, particularly one with access to customer data or core systems, is a good moment to review your existing insurance arrangements.
Speak with an Insurance Adviser
Every business’s risk exposure is different. If you’d like to understand how your use of AI tools impacts your risk exposure and insurance program, contact the Clear Insurance team and ask about our no-obligation risk and insurance review.
General Advice Warning: This advice is general and does not take into account your objectives, financial situation or needs. You should consider whether the advice is appropriate for you and your personal circumstances. Before you make any decision about whether to acquire a certain product, you should obtain and read the relevant product disclosure statement.
Clear Insurance Pty Ltd. ABN. 41 601 916 689. AFSL No. 548953.
Last updated: 25 June 2026
