Essential ESG: Episode 33 – AI in employment and ESG accountability

Sophie Parr, ESG Consultant, Corrs Chambers Westgarth 

Rachael Malarowski, Special Counsel, Corrs Chambers Westgarth

Sophie: Welcome to another episode of Essential ESG, coming to you today from Brisbane on the lands of the Turrbal and Jagera people. My name is Sophie Parr and I'm the Responsible Business and ESG Consultant at Corrs Chambers Westgarth. It's my pleasure to welcome to the podcast today, Rachael Malarowski, Special Counsel in Corrs Employment, Labour and Safety Team. Thank you for joining us today.

Rachael: Thank you, Sophie. It's lovely to join.

Sophie: Today we will be discussing AI in employment and ESG accountability, looking at the governance implications of deploying AI in workplace decision-making, the risks around algorithmic bias, and how Australia's evolving regulatory landscape is shaping the way organisations approach these challenges, and ultimately ensure businesses are employing responsible AI practices. So Rachael, while AI is the current hot button topic that's dominating headlines, people may not be as familiar with the current use of AI in an employment context. As such, can you give us a brief overview of how widespread the use of AI tools are in the employment space and what kind of decisions are currently being automated or augmented?

Rachael: Look, Sophie, it will be of no surprise to you or our listeners, I think, that AI is really quickly becoming the norm in the employment context in Australia. And the key areas for its adoption by employers that we're seeing are really throughout the employment life cycle. So starting in recruitment through to employee performance management and development, employee monitoring, and more broadly things like workforce planning and legal compliance. There's varying reported figures about adoption of AI, but I think it's generally accepted that use of AI in recruitment and onboarding practices in Australia is already really widespread. Some of the most common applications in that context are perhaps the ones that you would think of, so automated resume screening, candidate scoring and shortlisting, and even online pre-screening and automated interviews, which I know get mixed reviews from candidates. But within the workforce, AI is increasingly being used to manage rostering, work allocation arrangements and can be used for workforce monitoring as well. So monitoring things such as keystrokes and application usage and email volumes, monitoring remote workstations and also identifying health and safety hazards. But I think like any accelerated change and technological adoption, the law and governance is still catching up and there are some really key risks in each of the areas of AI adoption that it's important companies and companies’ leaders are really alive to.

Sophie: Yes, just picking up actually something you mentioned on recruitment as one of the key areas AI is being readily deployed - because this brings up a host of ESG related risks such as possible discrimination, oversight, accountability issues - just interested to hear your perspective on the risks of algorithmic bias in AI-powered recruitment tools? So things like the resume screening, video interview analysis.

Rachael: Yes, absolutely, Sophie. I think it was perhaps initially hoped that AI would be a tool to end all human bias and discrimination in hiring processes. But certainly the rhetoric thus far is that it actually creates a real risk to the contrary. Principally, that's because of algorithmic bias, which you mentioned. And in the context of AI powered recruitment tools, that refers to the tendency of automated systems to produce outcomes that systemically disadvantage candidates on the basis of characteristics such as gender, race, age, disability, or other protected characteristics and attributes, even when those systems perhaps appear neutral or were not designed with any discriminatory intent. And it arises primarily because AI recruitment tools are typically trained on historical data, which is often drawn from an organisation's past hiring decisions or current workforce makeup. And if those historical patterns reflect existing human biases, for example, perhaps a tendency to hire men in senior technical roles or to favour candidates from a particular university, then the algorithm learns and replicates those patterns at scale. So the bias becomes embedded in the model's logic, even if that wasn't the intention. Video interview scoring is an interesting example because that analyses the tone, facial expressions, speech patterns, which could foreseeably disadvantage candidates with accents, with speech impediments, neurodivergent communication styles, or even physical disabilities. Another example might be AI systems that reward linear career progression, because that could disproportionately disadvantage parents returning from parental leave, people with disabilities who have had interrupted careers, or even older workers who have changed fields over time. And that legal risk, that discrimination risk, is really compounded by the reverse onus under the general protections provisions of the Fair Work Act, which I think are worth highlighting because in those disputes, once adverse action and a protected attribute are established, it then falls to the employer to prove that the attribute was not an operable reason for the decision that was made. And where a decision was made or informed by an opaque AI algorithm, discharging that burden is foreseeably going to be really difficult for employers. So the core concern I think is that AI recruitment tools can encode and amplify human biases at speed and at scale and that can really produce discriminatory outcomes that are harder to detect, harder to explain and harder to defend than decisions that are made by a human hiring manager while the legal liability still continues to squarely sit with the employer. A difficulty at the moment is proving that discrimination occurred during a recruitment process. However, it's important to note that from December this year, there's changes coming into play in the Privacy Act that will require all covered entities to disclose the use of automated decision-making in their privacy policies. So that includes the kind of personal information used and the kinds of decisions being made by computer programs. That's going to make it easier for aggrieved applicants to identify that AI was used during the recruitment process, which could then lead to more discrimination claims and more exposure for employers using AI in their recruitment practices.

Sophie: Interesting. Interesting that those changes will be coming in soon as well. Thinking about some of the risks actually that you've talked about practically as well, what should organisations actually be doing to identify and mitigate these risks now?

Rachael: Sophie, in terms of what employers can do to manage these risks, I think the key factor is going to be having human input in the loop systems. AI should principally be used to assist, but in an employment context, it can't replace human judgment and humans should retain oversight in hiring decisions. On top of that, employers looking to use AI as a recruitment mechanism or as part of their recruitment process should be conducting impact assessments. So looking at what data their system uses, how it works and where a bias might arise. They should also be thinking about transparency by vendors providing these systems and what appropriate safeguards they could put in place, including in their contracts with vendors. Then also regularly monitoring outcomes to check for any disproportionate exclusion of protected groups. From a practical perspective as well, there's some really helpful resources for employers out there. For example, the Australian Human Rights Commission has published an AI in recruitment compliance checklist, which employers can use to help identify and mitigate risks of AI decision-making processes. And the Diversity Council of Australia also has some really helpful research on how to use AI recruitment tools to help rather than harm diversity in organisations.

Sophie: You also briefly mentioned earlier about AI being used as a tool to identify health and safety hazards. Can you expand on that?

Rachael: Yes, Sophie, absolutely. So AI is certainly a tool that can help employers identify, eliminate and minimise workplace health and safety risks. For example, using AI computer systems to identify hazards in remote work sites is something that has already been adopted at a number of mine sites in Queensland. And we're also seeing regulators like Resources Safety and Health Queensland recommending the use of AI-enabled safety technology. But there's also another workplace health and safety angle when rolling out AI. And that is that AI itself in the workplace can create new workplace health and safety risks. Safe Work Australia has expressly called out AI as a risk area for psychosocial harm, particularly where systems are introduced without adequate training, responsibility shift or workers’ experience, reduced autonomy or job security as a result of the AI. So AI can have clear benefits for work health and safety and is being used in that way, but it also comes with its own risks that need to be assessed and managed just as you would any other physical safety risk.

Sophie: Given some of those work health safety risks arising in that context, what's your view on what good governance looks like when an organisation decides to deploy AI responsibly?

Rachael: It's a great question, Sophie. And when thinking about governance from a workplace health and safety perspective, really the starting point is a person conducting a business or undertaking or PCBU's primary legal duty, which is to ensure so far as reasonably practicable the health and safety of workers and any other person who might be affected by the work. And that extends to both physical and psychosocial health risks, both in equal measure. So in managing AI implementation from a governance perspective, employers should apply the same risk management approach they would for any other workplace hazard. So identify, eliminate or minimize and control risks through a documented system, then monitor, evaluate and reassess the risks and controls periodically. Beyond that, there's a number of things they can do. Workforce consultation is a really important consideration to be alive to. Of course, there are consultation obligations under workplace health and safety laws which require consultation with both workers and other duty holders in managing workplace health and safety risks. But also workplaces covered by an award or an enterprise agreement will have mandated consultation obligations when major workplace changes occur. And introducing AI tools can trigger those obligations. Unions have already flagged that this is a priority for them and that employers with consultation obligations might find themselves in dispute proceedings if they fail to consult on an AI rollout that amounts to a major change.

Sophie: Just pivoting now to the regulatory framework for AI because it has been moving very rapidly over the past few months. Can you walk us through some of those recent Australian government announcements and policy developments around AI and employment and what these specifically mean for employers in practice?

Rachael: Yes, absolutely. As you say, this is a very quickly evolving area and there's lots been happening. We touched earlier, Sophie, on the Privacy Act changes, which are going to require disclosure of the use of automated decision-making processes in privacy policies from December this year. There is an employee records exemption, which provides some relief for private sector employers in respect of managing employees. But that exemption doesn't extend to recruitment or to the management of contractors and labour hire workers. So all APP entities will be required to disclose automated decision-making in those contexts. I think the big one that people will have seen in the headlines was in July this year, the Prime Minister announced the government's increased focus on AI regulation, including in the workforce sphere. And really the central theme of that was that employers will be held to a high standard in terms of protecting workers' health and ensuring that they continue to have secure and fulfilling work. In mid-September, Anthropic signed its first Australian data centre agreement, which is a lease at a $32 billion site being built near Dalby in Queensland, targeted for 2027. That kind of infrastructure investment underscores really the practical urgency of getting the regulatory framework right. Before the PM’s speech back in April this year, the Workplace Relations Minister Amanda Rishworth announced the establishment of the AI Employment and Workplaces Forum, which is a body that brings together the ACTU, the Business Council of Australia, Australian Industry Group and the Australian Chamber of Commerce and Industry alongside the government. And the forum's remit is really broad. It's essentially to chart a path on how AI is adopted in workplaces, ensure that there are productivity gains which is shared with workers and to test whether current laws are fit for purpose. Most recently, though, in August this year, a joint select committee on artificial intelligence was appointed by both houses of Parliament, and it has a wide remit as well to examine the risks and opportunities of AI, review the adequacy of existing laws, including copyright and IP, and to consider issues such as national security, data sovereignty, consumer protection and deepfakes. That report's due in November 2026. So we may see further legislative signalling before the end of this year. One development that might surprise listeners perhaps is actually the impact AI is having on employer-employee disputes and in particular the Fair Work Commission's own caseload. The Commission's total workload has increased over 70% in three years. And that increase has principally been attributed to the growing use of AI by employees or potential litigants. A research report commissioned by the Fair Work Commission found that approximately 40% of surveyed applicants used generative AI to prepare their claims, with ChatGPT being used by 77% of those users. The report also found that, to quote “sycophantic or hallucinatory AI outputs were reinforcing applicants' positions and inflating expectations, which in effect makes them more reluctant to settle and makes disputes last longer”. In response, the Fair Work Commission has introduced a new guidance note which took effect in October this year, which requires parties to disclose where generative AI has been used in Commission documents and verify all content for accuracy, etcetera. But really the practical effect of this is that there is a real increase in claims that employers are dealing with and sometimes an unrealistic expectation of outcomes for employees who are receiving legal advice from AI. To give you a good practical example, I think, Sophie, of how AI is complicating employer-employee relations, in late July this year, as part of a bullying complaint that was before the Fair Work Commission, the Commission actually ordered both the employee and their Queensland employer to stop using AI in all future correspondence with each other because AI-generated communications were found to be escalating the dispute rather than resolving it. So it's not just about claims being filed and more employees bringing disputes, but AI in some contexts is actually actively changing the dynamics of employer-employee communications.

Sophie: So you've spoken about that from a federal level, but what about at a state level? Have there been any government announcements and policy developments around AI and employment?

Rachael: Yes, absolutely, Sophie. From a workplace health and safety perspective, New South Wales is at the forefront having passed legislation in February this year. It’s called the Work Health and Safety Amendment Digital Work Systems Act, a long one, but in effect it introduced a new duty of care requiring PCBUs to ensure that digital work systems which includes AI, algorithms, and automated decision-making, don't put workers' health and safety at risk. The Act specifically addresses risks from excessive workloads, unreasonable performance metrics, excessive surveillance, and also what we spoke about earlier around discriminatory decision-making. Workplace health and safety entry permit holders will also have a right to inspect digital work systems. And those new duties haven't commenced yet, but they're awaiting proclamation and also Safe Work NSW guidelines. In Victoria, there's also been some developments in terms of workplace surveillance reforms. The former Labor government announced what it described at the time as the toughest workplace surveillance laws in the country, which were said to be technology neutral requirements to consult, notify and prove surveillance is reasonable and proportionate. The position on those reforms is currently unclear following a change in leadership, but they remain on the agenda for the November 2026 election. So it's certainly a watch this space.

It's also worth noting that South Australia has announced a Royal Commission into artificial intelligence, which is expected to commence in October this year with a final report due by July next year, 2027. And the scope of that extends beyond employment which really signals the breadth of scrutiny AI is now receiving at a state level. And one of the most practical developments I think though is happening in bargaining rooms rather than parliament. Enterprise agreements are increasingly regulating AI at work with clauses on mandatory pre-implementation consultation limits on AI only decision-making, surveillance restrictions and workforce transition protections all being considered as part of bargaining. And for employers negotiating enterprise agreements, AI is now firmly on the bargaining agenda.

Sophie: So how would you say then Australia's approach compares internationally?

Rachael: Well, Sophie, really interestingly, the EU is leading on international standards and has already passed a dedicated AI Act, which treats many employment use cases of AI as high risk. So that includes recruitment, promotion and termination decisions, task allocation, and performance monitoring are all classified as potentially high-risk uses of AI. Some obligations already apply under that Act, such as a ban on workplace emotion recognition which came into force in February 2025. However, the deployer obligations which were due to come into effect in August this year, 2026, have been delayed and simplified by the 2026 Digital Omnibus, and the high-risk requirements will now apply from December 2027. Providers will then need to comply with extensive requirements under the AI Act, including risk management systems, bias controls on data, and built-in human oversight. Employers using these tools will have to assign human oversight, inform workers and their representatives, and in some cases, mainly public bodies, they have to carry out fundamental rights impact assessments. So as Australia's regulatory regime is yet to be developed, that's certainly something that Australian employers can look to.

More generally, engagement with higher risk AI applications in a workforce context needs to correlate with stronger responsible AI capabilities. Mature organisations in this space leverage better governance frameworks to safely deploy complex use cases for AI. So effectively, they can be bolder in deploying AI in high-risk employment contexts because they've invested in the governance infrastructure to do so responsibly. They have clear policies, accountability structures, bias testing and so on. Less mature organisations should appropriately limit exposure to challenging applications of AI.

Organisations that haven't yet built strong governance capabilities should sensibly limit themselves to lower risk, simpler AI applications. Generally, that means not deploying AI for things like recruitment screening or disciplinary decisions because they know or at least sense that they don't yet have the guardrails in place to do that safely. So ultimately, organisations getting the most value from AI in employment are the ones that have invested in governance first. And those that haven't built the frameworks yet should be cautious about where they deploy it until they build that capability.

Sophie: And finally, in such a rapidly evolving space, where do you see this heading over the next few years? For instance, what might the next wave of regulatory intervention look like? Also, what should organisations be thinking about in terms of long-term positioning at that intersection of AI employment and ESG?

Rachael: Yes, Sophie, this area is developing daily at this point and we're likely to see significant change in the regulatory landscape. Currently we have a piecemeal approach with different states legislating separately, but the national agenda has shifted towards a more cohesive framework. Key areas of focus are going to remain surveillance, work allocation and recruitment, which we've touched on today in a bit of detail. In terms of the concrete timeline, we already know we have the Joint Select Committee reporting November this year, the South Australian Royal Commission I mentioned, which runs from October through to mid next year, and the Prime Minister has actually indicated that supporting legislation for national AI standards could be before Parliament early next year in 2027. So the pace of change is genuinely accelerating. My advice, I think, to organisations is really to get on the front foot with their AI governance, develop policies and strategic governance plans now, rather than trying to backfill once legislation is introduced. It's far easier to manage when you're building governance frameworks alongside the implementation or rollout of new technology.

Sophie: Thank you so much for sharing your insights today, Rachael. It's been so wonderful to cover such a topical issue and we look forward to speaking with you soon.

Rachael: Thanks for having me, Sophie.