Public practice warning: AI boom opens the door to cyber crime

This article was current at the time of publication.

In the ever-evolving landscape of accounting, artificial intelligence (AI) has emerged as both a powerful ally and a potential threat. 

Financial and insurance services across Australia and New Zealand are coming face to face with the stark reality that AI is revolutionising not only how we protect sensitive financial data but is also arming cyber criminals with sophisticated new tools. 

Email phishing, text phishing and robotic-sounding voice calls are still the most common cyber scams being used by criminals to attack businesses rich with a treasure trove of data and information, such as accounting. 

But the AI boom has now spawned AI generated content that can mimic legitimate communications more effectively and clone your voice and image from publicly available audio and videos to scam anyone from family and friends to company executives. 

This deep fake technology is already causing large-scale damage and chaos. 

Scams are all too real

Graham Chee FCPA, managing director of cybersecurity risk consultancy BCyber, says AI is a double-edged sword; it’s not only enhancing security measures but also potentially opening new avenues for cybercriminals. 

“AI's capabilities in threat detection and response are truly remarkable. However, the same technology that protects us can also be weaponised against us. I've seen AI-crafted emails that could fool even the most cautious accountants. 

“The pace of AI development is outstripping our ability to regulate it. We're in somewhat uncharted territory.”

Chee advocates investing in AI-powered security solutions: “Fight fire with fire. This is not science fiction – this is happening now.”

Reporting down in New Zealand 

Cyber criminals’ ability to manipulate AI means they’re finding new targets all the time.

In New Zealand, the National Cyber Security Centre (NCSC), notes that Maori and Pacific Islanders were previously not common targets of traditional phishing scams. 

Now, thanks to AI’s Large Language Models technology, these “Regional, Culturally Significant (RCS) communities” are targets.

“These are languages spoken by smaller groups of people but in specific places, for example, Maori, Samoan or Icelandic,” says a spokesperson. 

“AI tools can be used to create more realistic phishing emails and websites, and in a variety of languages. AI is better at simulating subtle cultural rules by native speakers. 

“AI is also a relatively cheap tool, which means criminals can target even more people and with greater success.”

Phishing is still the most common cyber scam but criminals have gone beyond trying to get information out of a business; “scammers are also impersonating financial organisations”.

New Zealanders lost NZ$3.6 million to cyber scams between April and June this year. While attacks have increased, the reporting of these crimes has dropped off significantly.

“We’re investigating, but the cause is not clear,” says an NCSC spokesperson. “We are fairly sure that businesses are not hiding reports out of embarrassment or reputational reasons. It may be because they have reported to their bank or police or another agency.”

Premiums stable but small businesses folding

New Zealand businesses are investing in cyber insurance but not as much as other territories, and lag behind Australia in terms of general policy uptake, says Jono Soo, Head of Cyber Specialty at Marsh Ltd* in Auckland.

“There are regulatory and legislative factors which are driving businesses to manage cyber risk seriously in Australia, and we do foresee that trend coming across to New Zealand in the next three to five years if the government chooses to follow the Australian legislative changes.”

AI hasn’t shifted the claims or loss landscape from a cyber insurance perspective according to Soo. Most businesses impacted by ransomware or cyber extortion threats generally choose not to pay and go down the recovery route. 

“The costs of responding to an incident, regardless of paying the ransom, can easily cost six-figure sums even for minor incidents. A major or significant incident can easily cost millions to recover from.” Businesses, especially small ones, have folded. 

The best action, says Soo, is to “continue to educate employees and implement processes and technology that will prevent, detect and respond to intrusions”.

“Treating AI as a natural evolution of digital risk allows risk managers and internal IT and security teams to conceptualise and mitigate the risk in simpler, more manageable terms.”

Be sceptical and verify everything 

Social media is being blamed for making things easy for cyber criminals. 

Tyler J Wise, Director and Lead Investigator and Trainer at Cyberwise, says you don’t need to stay off media channels, but you do need to be “sceptical and verify everything”. 

What about company websites?

Wise says: “When done well, they can be a genuine marketing difference, however, if not monitored and maintained, the more code a website has, the more risk it may have as well for an attack. For organisations that are not going to be able to dedicate resources to constant website maintenance and engagement, a static page would be a wiser option to start with.”

Don’t be a sucker for freebies

Wise adds: “Anything free is a risk, especially if you are not given the option to stop training the model. We need to ensure we maintain ownership of the data and that it is only used for our purposes.

“Take a tempered approach and ask, 'Do you need to be first to use this, and if so, what are the business gains?' A lot of the time, it is excitement that drives the adoption of AI, and the reality afterwards is that it didn't drive as much efficiency as hoped. Wait until there has been research, external verification and wider use, so some key metrics can be made available, along with the all-important security commentary.”

Do you have this checklist?

Wise recommends the following:

1. Have strong internal governance controls, e.g. policies and procedures for use and inputs when deploying AI tools.
2. Conduct internal audits on the model, and test to ensure outputs being delivered are accurate and current.
3. Ensure information being entered into the model is isolated and not being used to train the model. Most important for organisations that use the model for sensitive engagements, or where staff may inadvertently reshare information. 
4. Remove the 'history' from the tool. Especially with Large Language Models, where old queries should not be able to be retrieved.
5. Ensure AI tools have standard cybersecurity measures in place, such as strong password protection, and multi-factor authentication.

Bring in a ‘penetration tester’

Miranda Riddell, Manager of Malware Security, says accounting is a high-stake business with extremely, sensitive and lucrative personal identifiable information, so using an “ethical hacker” or a “penetration tester” to identify vulnerabilities should be seen as a standard business expense. 

“Penetration testing is often begrudgingly perceived as an additional expense, but a security breach would be significantly more costly to the business - both financially and reputationally.

“Practices may also have legal obligations to test their systems regularly, so annual penetration testing may not only be useful, but required.

“Arguably, just as important as penetration testing, are social engineering assessments, such as phishing exercises to test employee awareness and resilience against scams. These can be part of your penetration test, or their own activity. 

“Your technological environment is only as secure as its users, and humans are often the weakest link.”

In November CPA Australia is hosting a cybersecurity webinar for members. For more information, access here. 

Disclaimer: This article should not be considered as legal, tax or financial advice and may not reflect the views and opinions of CPA Australia.  CPA Australia does not warrant or make representations as to the accuracy, completeness, suitability or fitness for purpose of this publication and disclaims all liability and responsibility for any acts or omissions made in
reliance of this publication. Readers should seek their own professional advice that takes into account their own personal circumstances.