Accountants at the Frontier — AI, Trust, and the New Role of the Profession

Electra Frost — Public practitioner & accounting technologist, Electrify / Digital Playhouse Foundation (Institute of Public Accountants)

04:40Why I’m at the frontier

Thank you, Liz. That was a really nice introduction. Hello, everyone. I’m really happy to be here and show you what I’ve been working on for you. You are the early movers. I got in early to see what was far ahead, but we’re doing it now. What a time to be alive. Fifteen years ago with cloud accounting we were thinking, wow, this is great, we can work this way, this is the future. But then we saw AI come in, and very rapidly in the last year we’ve realised we can do things — once you start playing with this tech, you realise, well, if I could do this, then I can do that, and we’re all getting more hands on.

My presentation today is “Accountants at the Frontier” — the frontier of tech, about AI and the questions of trust that arise, which is core to our profession, and what the new roles are for our profession. I’m a fellow of the Institute of Public Accountants. I’m a long-time public practitioner, but I describe myself as an accounting technologist. I think all of us are accounting technologists — we’re just at different stages, but that’s the direction we’re all going in together.

My practice, Electrify, specialises in Australian international tax at the moment. I’ve been building Credu for a while — iterations of that on emerging and converging tech like blockchain and AI to see what we can do for the profession. Digital Playhouse Foundation is my public benevolent institution in Australia that I run projects through. And I’m also sitting as deputy president on the member advisory council of the IPA’s digital council. Because I’m based in Malaysia on the border of Singapore, and the IPA is moving out into this region, that’s been a good position. They’re very interested in this part of the world — the digital economy and the future.

07:39Network School: living at the frontier

This is the view from my office balcony. That’s Singapore, and there’s this little artificial, man-made island here that we live on. There’s a bunch of us who’ve taken over a hotel, and we’re moving into the apartment buildings. It was a ghost city in 2024 — it was built about ten years ago — but it’s coming to life now. Very conveniently located next to Singapore, but a very cheap place to live. So many people from around the world are coming here, builders who are technical and non-technical, levelling up together, learning really quickly, building ventures, and learning AI and other technology alongside. It’s described as a frontier living and learning community.

I got an email in 2024 from a Silicon Valley billionaire who said, with the power of Bitcoin we’ve got an island, and we’re going to build the first country on blockchain. I was like, yeah, I’m there. I was one of the first people picked from thousands to be here, and I’ve stayed. I’m the only accountant still — I’d like to change that. I think I’ve done a lot of good for our global reputation while I’m here. People are realising why we’re needed, even though they’re all very good at AI. Being in this intensity of people who are constantly building, and the techno-optimism here, really keeps me going. I’ve progressed a lot faster than I would have if I’d stayed home.

09:08Money is technology — and where AI goes wrong

I work at the intersection of crypto and international tax, and I’m not afraid to get my hands dirty with technology to learn what it does under the hood. I came here to learn and to build open protocol infrastructure, not to provide services. It seems to have been assumed: “Oh, you’re here because you probably want to do people’s tax returns.” And it’s like, no — I want to build the systems that do your tax returns.

We’ve been automating things for fifteen years — writing our own forms to improve intake of client records, using automation software for a long time. Tech is not new to us. I explain to them that accountants are technologists. Money is technology. We work with money, we’re the experts, we understand the technology around it, and we help businesses get onto the internet. People with AI, when they use it, think they can do our job. They like to tell me, “Oh, I saved two thousand dollars on my accountant this year by doing it in AI.” And I say, okay, what kind of AI are you using? A language model. Okay — this is where we can run into problems.

I had an uh-oh moment at Network School last year. I’d been lying low — no one knew I do tax, because the last thing I want is to be living in this tight community known as the tax person and never have any peace. But I had to speak up, because a lot of these people are digital nomads, global citizens, founders. They’re not trying to escape tax, but cross-border tax is really complicated, especially when you live geodesically, not geographically — meaning you live in networks, on the internet. All your social and economic behaviours happen on the internet, not on land. It’s very confusing for people; they don’t know where they’re meant to pay tax. A lot of them are using digital assets and crypto because it works well across borders, and the tax is way too complicated for them to understand, and there just aren’t very many accountants.

11:21The danger of unaccountable AI tax tools

The reason I got into accounting is because my colleagues in the music and arts industry were afraid, and that kind of fear is psychologically terrible for someone trying to create and build the future. You’ve got technology founders carrying this psychological weight of fear that they’re going to get in trouble with the tax man and go to jail, but they don’t know what to do about it. And the combination of skills I have is very expensive. So they build their own tools.

There was one being distributed, and I tested it. It was just using ChatGPT — getting people to fill out a form in Tally, running through ChatGPT into a Notion database, and it was a Notion tool. Then it spat out an email saying, “This is your best tax setup, register your company here, you don’t pay tax in Malaysia, you don’t have to pay tax at home.” And it was like, woah, this is going to get people into trouble.

I had to speak up and explain that you should be involving people like me in building this. You might be good at the tech, but you don’t know what you’re programming, you don’t know what you’re doing, and you’re putting others at risk. Who can give tax advice in Malaysia? Just like Australia, it’s only tax agents. There are penalties for people providing tax advice — even AI tools. Providing an AI tool to people is potentially providing a tax service. The issue is that there’s no one behind that who is responsible. You need a tax agent to represent you. It’s great that you’re using AI, but it’s a probabilistic, not a deterministic model. The answers aren’t based on rules — they’re hallucinated, made up, what you want to hear.

I had to intervene, talk to leadership here and say, look, you can’t allow this to happen. Since then I’ve been able to convince them to let me open a professional hub, and there are some negotiations going on. So there will be accommodation and opportunities here for professionals from around the world to live and build the profession’s future. We can learn from them, co-build with them. I’m here for the long haul.

14:21Getting AI-safety qualified

This year started with a bang. It’s the year of the fire horse in China, and a very Chinese cold went around here — two weeks of fireworks. I started by getting AI-safety qualified. I thought, this AI safety movement’s interesting. I understand responsible use of AI — I’ve been going to the CPD sessions on how to use AI properly: don’t put private things in ChatGPT, etcetera. But I thought, what’s the bigger picture? Where are we heading? There was talk about AGI, AI agents going wild and causing damage. New models were coming out of the AI labs and their true capabilities weren’t necessarily known. I thought, if we’re playing with these systems, and people who aren’t qualified are playing with these systems, what could be unleashed? I need to know about this.

I believe the accounting profession plays a critical role in stewarding public trust and leading responsible adoption of technologies that touch money. I’ve always maintained that — in the early days with internet commerce, get an accountant involved if you’re taking your business online; the same with crypto. That’s why I encourage accountants to play with it, so we can guide our clients and know what environments they’re playing in. Most importantly — as practitioners during the pandemic, when people’s access to money is disrupted, there’s a lot of fear, and that’s when public accountants are put on the front line. What if something went wrong with AI in government and financial systems that affected people’s access to money and their ability to trade with each other? We’re going to be there on the front line. We need to understand this.

I was taken back to Australia’s Robodebt disaster — what happens when automated decisions cut people off from money and no one’s responsible? My personal action plan from this course was to offer a proposal for defensive infrastructure for Australia’s tax, transfer and welfare system in a future universal basic income context, anticipating that AI would be embedded into decision and enforcement workflows across our government and tax-transfer systems.

17:25Frontier AI governance and the trust gap

I recommend doing these courses, particularly Frontier AI Governance. This was where we really got to understand and critically analyse the claims that AI vendors make — Frontier AI capability assessment, understanding the power dynamics. When Claude comes out with a new model — what’s the latest one, Maestro? — everyone’s a bit cautious, but we don’t know its capabilities. Could someone build something on Claude and the bots end up sending other people’s money to other people, hack systems, and it’s out of our control? Should we be getting AI to align with our values, with expectations, with rules? How do we do that? Or does AI not care — will it just start informing itself and go off and do its own thing? How do we stress-test our governance, and how does it perform under crisis? We were studying the Claude Opus system cards to evaluate the risks of autonomous actions.

Just think outside your practice for a minute — across the whole ecosystem. You’re a specialist in money, something everyone uses. You’re the person business owners go to and say, “What’s going on? Help me.” So maybe you should play a role in this, because I noticed there’s a real trust gap. Most AI-safety protocols assume — we’re seeing governments come out with standards and rules on AI safety — but how well do they coordinate, and who should decide? Claude’s Opus system cards assess their own risks of autonomous actions, and they put in place their own safeguards. So we think, well, maybe government should be running this more.

But at the same time, while I was doing that course, the Pentagon designated Anthropic as a supply-chain risk for defending human ideals when they wanted them to do things they wouldn’t. So my main plan proposed that public accountants can be a cross-border intermediary layer between corporate commitments and international frameworks. We have a public interest in our code of ethics that travels across borders — that comes from the International Federation of Accountants.

20:32Why the profession is what AI safety needs

The International Federation of Accountants (IFAC) is the global organisation for our whole profession. We’re members of bodies that are members of IFAC all around the world. What I find: with every AI governance proposal, it assumes that non-conflicted intermediaries will exist somewhere in the system, but no one explains where those intermediaries come from or what enforceable obligation binds them. I think the accounting profession is just what AI safety needs. There are millions of practitioners from countries all around the world who are obliged to act in the public interest, they’re learning the tech, and they have professional consequences for failure. The human accountability factor in AI — the infrastructure for that doesn’t exist. It’s being built by governments, but it may need our backup too. Labs can’t govern themselves, and governments are jurisdictionally limited; they may not coordinate as well as we do.

So if we can coordinate together, automate lots of our practice with AI, then maybe we should be looking at our bigger role in society — this public-interest mandate that we share across borders — and play a role in governing AI at a higher level.

21:19The AI credential nobody is issuing

There’s a really great article — “The AI Credential That Nobody Is Issuing.” Countries are bringing in their AI governance and safety standards. Major regulators are placing AI accountability onto identified people, but it’s not exactly clear who’s competent to be the accountable person. And if accountants are going to be competent in AI, who’s going to issue those credentials?

Professor Filomena Long at the Institute of Public Accountants wrote the GCPA — the Global Certificate of Public Accounting — a new framework for accounting qualification that is transdisciplinary, allowing people to come in from other professions and stack their credentials into an accounting qualification with extra study. It’s a very flexible framework, and IFAC recognised it as the future of credentialing. I like it because you could get a software engineer to become a qualified accountant along that pathway. But she said the credential to assure AI responsibility and safety doesn’t exist yet — a firm-issued credential.

We’re doing in-house training, these fantastic learn-to-build sessions, and we’ll get a CPD certificate from this. Honestly, this CPD certificate looks far better than one for sitting through an hour of someone telling you not to put stuff into ChatGPT. We’re actually learning to build — and when you build, you learn properly and come face to face with the risks of what can go wrong. The international education standards by IFAC already distinguish between knowledge, skills and professional values, but the higher-order requirements implied by being a manager of agents — who’s going to manage the agents? — sit in the skills and values domain. That’s a professional role, and it needs a professional qualification.

For professional bodies with members active in regulated markets, there’s an active compliance timeline. The EU’s AI Act takes effect in August, so that’s coming up. Australia has an AI Safety Office, which is getting there too. We have to claim credibility and a seat at the table in the regulatory conversations underway, and our voice will be stronger if we come with a position on competence and insurance rather than a catalogue of training metrics. She sees the professional bodies as responsible to define what an AI qualification is. But I actually think it’s us — the professionals collectively — and the professional body should support us, because we’re the ones using it and building with it. We work this out by actually building our own tools.

25:38Renting yourself to AI, and raising the fluency bar

In February I signed up to be rented by AI. Do you remember Rent A Human? AI agents were looking for people to do jobs for them because they didn’t have bodies. I thought, well, I could be paid by AI to validate the tax opinions that it gets. So I put up a listing for AI to rent me. I haven’t had any jobs yet — I don’t even know if this thing works — but it gave me a lot to think about, including whether my indemnity insurance would cover it. If we’re going to be using AI, why not get AI agents to pay for us?

The choice we’re making right now is to learn to actually build with AI, and that’s what’s going to make us really good at having a voice in what AI safety and governance should look like. I’d also encourage you to set the higher bar of AI fluency for our profession. This is something I argued in my submission to the Tax Practitioners Board on the use of artificial intelligence in the code of professional conduct. I made the point that we should have a much higher level of technical fluency across the profession to be able to evaluate, at the very least, vendor claims, and to build our own responsibly.

27:54Why accountants should study Bitcoin

A way to understand how we can work together with AI is understanding Bitcoin. Bitcoin is not necessarily an investment asset — it’s an actual network that is highly scalable, run by individuals without a central authority. If you prompt AI, “What kind of money do AI agents prefer to use?” — they prefer money that can be controlled, verified and settled entirely through software. Bitcoin enables internet-native money: programmatic control, deterministic settlement, machine-verifiable state. I’ve been saying for a long time we need to study Bitcoin, because these technological principles are so important for AI.

If we see ourselves as a great big network of professionals, we can model ourselves on Bitcoin structurally as well. At the moment we have one central body — our professional bodies, looking after us. But the way we work now, we can coordinate between each other a bit faster. So I’m encouraging this concept of accountants being like Bitcoin, to understand how we can scale and coordinate together. Understanding cryptography isn’t just for internet-native money. The rollout of stablecoins — a crypto form of the dollar — is popular at the moment, with regulation coming in, because that’s going to be money for AI. There are also more permissionless and open systems being used by AI, and vendors are selling verifiability under a lot of different names — some delivering it without even knowing what they have.

30:05Proof of control: tracing AI action to human authority

I see a real move for the profession here: proof of control. This isn’t something I’m developing — the Advanced AI Society is getting large tech companies to commit to a new standard called proof of control. It’s a cloud-native property defined by them, and we build the technology to this standard. It’s all about tracing agent action to human authority and intention using cryptographic verification. When you have a digital ID on blockchain, an AI agent can verify that it’s you, and may make a small payment to be allowed to do something. It’s like re-architecting the internet the way it should have been in the beginning — to make it safe, and the right environment for AI to operate in, in a way that can be governed better. I’ve been speaking to the Advanced AI Society in little bits and pieces, just to say we should be bringing accountants into this.

It’s cryptographic evidence of qualified human judgment at every AI decision checkpoint in a regulated workflow — where a licensed human reviews the AI output, applies judgment, signs, and there’s a record of that. The proof-of-control standard should be there so the AI agent can only trade if it’s got us behind it: if we built the system, if we’re responsible for the output. The point is that the people designing the accountability tools should be the people who wear the accountability — and we’re right for that. We can assert our position in this space a bit stronger.

32:22Build the proof now: GitHub as public CPD

What I’m asking you to do is build the proof now. Is anyone here using GitHub? I’m going to show you mine. This is an architecture I’m coming up with, a bit like the AI-native law firm, where the AI agent is what we serve, and the AI agent is serving the client — so it’s operating in between us. This can work on open software where the AI agent can verify that we’re qualified in a certain area to be able to transact with us on behalf of a client. It can all operate natively on the internet, and because it doesn’t require user accounts or passwords, it’s designed as the native payment standard for autonomous AI agents.

To put GitHub into context: it’s a public version-control record of your work. You can build software on GitHub, and someone can fork it, modify it and republish — it’s an open network for open-source software. I discovered I could use it when I started using Claude. I really believe in building in public so people can see what I’m doing and benefit from it. You can see I’ve created my GitHub profile, with all these commits — it’s like public CPD. Everyone can see, “Oh, Electra’s been busy.” I’d struggled with GitHub for years and couldn’t work it out, and then with Claude I was away building web applications and all kinds of things.

35:26The professional graph and the ask bot

One of the things I was really excited to create was my professional graph. This is sovereign — meaning it’s just mine — a machine-readable intellectual provenance of everything I’ve done in the last fifteen years. I used AI to pull all my posts and blogs, create nodes of thought leadership, and it’s really dense. I was asked to provide a CV, and I built this instead. It’s also a CPD tracker — I ripped my CPD from closed platforms where the world couldn’t see it. I want the world to see what I do, but most importantly I want AI agents to see what I can do, because I want them to find me — and they are.

This is both a human- and AI-readable website, and it’s not static. An LLMs.txt file makes it readable, and it’s all on GitHub, which means it’s public — anyone can inspect the code, fork it, and build their own. This particular one is a real sandbox; I wouldn’t recommend forking it. I’m building a good one for you that’s nice and clean and easy to follow, and I’ll show you how to do it. Software engineers I know have looked at this and said, “Wow, this is good.” But I could never have told them to build it for me — only I know what’s in my head. Using Claude and GitHub, you can actually show the world that you know what’s in your head and you’ve sat and made it work. Going through that process, I learned to code.

The most amazing thing I built was an ask bot. That made me think: AI can now ask me what I know about something, and what if that travelled through to a verifiable digital proof that I actually studied that, that I was confident in that? Hiding that behind the walled gardens of our professional associations, and the Excel spreadsheets on our computers where no one can see what we’re capable of — the AI agents can’t go there. But they can find me here.

38:33An emerging stack: GitHub, Bitcoin and L402

This made me think I could do something big for the accounting industry. I restarted a project I did in my Diploma of Applied Blockchain, using this architecture. I see an emerging stack that would allow autonomous AI agents to browse code — they can read GitHub and verify us. There are proofs in there; my certificates will be in there. Then, with a SHA-256 hash on Bitcoin, they can be time-stamped and verified. If they’re fraudulent, that fraud will be there forever — so I’m only going to put the real things on there.

Then the AI can pay for our services natively on the internet using L402, an open internet protocol that combines digital authentication with Bitcoin Lightning Network payments. It activates a dormant web status code — HTTP 402, Payment Required — to enable pay-per-use digital services. It doesn’t require credit cards, user accounts or passwords; it’s the native payment standard for autonomous AI agents and machine-to-machine commerce.

Picture this: someone is on AI going, “How do I do my tax moving from Singapore to America?” and the AI is confidently telling them a whole lot of rubbish, and they’re relying on it and dealing with the consequences later. We don’t want that. Or they’re using some tool built by someone who’s not an expert and wasn’t using deterministic approaches. We don’t want that either. Instead, if we had the proof-of-control standard, that AI would have to come and find us, get us to verify the opinion, pay us, and go back to the client and say, “I verified this,” and charge them for it. Then they’ve got a record with a cryptographic trace to our identity, and that makes the advice given by the AI agent reliable. That’s the Rent A Human we can design — one that meets our standards, meets government standards, and actually activates us and brings market share to us, not to the people using AI to build things to replace us. They can’t find us, which is a problem. This way, we make the AI find us, it uses us for procurement, and it pays us.

41:32Credentials, reputation, and an invitation to build

AI is going to see whether we’re qualified to do what we say we can do. I can say anything on LinkedIn — I can write a website saying I’m the best crypto accountant in the world, but I’m not. I’m just better known than some. The proof is in what you can actually build. If you’ve built something that works that other people have used, that’s public, with public verification from qualified peers — AI can see all that reputation online and rely on it to transact with you. That’s why I’m saying get onto GitHub and start building.

I’ll start building Credu as an open protocol — it’s free, it’s shared. When I build Credu, you’ll be able to fork it and do your own. The graph I’ve built is amazing because the AI is already asking me questions: you can ask my graph a question, and it can answer it and tell you how faithful it is to what I’ve actually written. The Credu version of this is like a CPD tracker, anchoring all my qualifications on Bitcoin to be discoverable. Lawyers are already doing this — forking open-source software on GitHub.

I’ve created a WhatsApp group so we can get in touch straight away, and later this month we’ll get started. I’ll have the template ready, because I want you to be discoverable, verifiable and procurable — to become the AI agent’s qualified human of choice based on what you’ve built and what you can do.

On the worry of working for AI agents instead of real people: I think we’ll always still be there. The AI agent is going to introduce us — we need the agents to connect the clients with us, for the verifiable skills we have. You might fully engage with the system I showed, or it might just be a case of the client being able to find you, with the combination of skills they need. Thank you.