cd /news/ai-agents/letting-ai-near-the-ledger-five-cont… · home › topics › ai-agents › article
[ARTICLE · art-174299] src=dev.to ↗ pub= topic=ai-agents verified=true sentiment=↑ positive

Letting AI near the ledger: five controls Australian accountants should expect

Fedix outlined the design controls behind its accounting AI assistants, including a Copilot and the FediPilot desktop assistant, that keep a human in charge of every ledger change. The company said AI-requested changes wait for human approval, personal identifiers can be masked before content reaches model providers, and assistants inherit the signed-in user's permissions rather than a broad shared service account. Sensitive actions are logged centrally and ledger changes can be previewed, rolled back and undone.

recommendation

Require human approval for every AI-initiated ledger change, run assistants under the signed-in user's permissions instead of a shared service account, and mask identifiers before sending data to model providers.

more in playbook: Securing agents and AI tools
by read3 min views1 publishedNov 22, 2026

Originally published at https://home.fedix.ai/?utm_source=devto&utm_medium=article&utm_campaign=cold-start-2026-10&utm_content=pub-3ae03af2820a - canonical version there.

If you run an Australian practice, you have probably been asked the same question more than once: "Can we let an AI assistant touch the ledger?" The honest engineering answer is "only if a person stays in charge of every change." This post walks through the design principles we use at Fedix, and what each one looks like for the accountant using it. The most important rule is simple. Changes requested by the Copilot or a connected assistant wait for a person to approve or reject them.

The same thinking applies to Xero. Every change Fedix would make in Xero is confirmed by a person in a confirmation dialog first.

For you, that means the assistant can do the legwork, but nothing lands in a client's books until someone on your team says yes. Accountants handle names, tax file numbers and other identifiers all day. Personal identifiers can be masked in each Copilot session before content is sent to AI model providers.

If your practice has a policy about what client detail may be sent to a third-party model, this is the control to ask about. The details of what is masked and how are worth checking against our security page: https://home.fedix.ai/security?utm_source=devto&utm_medium=article&utm_campaign=cold-start-2026-10&utm_content=pub-3ae03af2820a A common failure of AI integrations is a shared service account with broad access. We took a different route. The FediPilot desktop assistant uses Fedix as the person who signed in, with the same permissions.

So if a junior cannot see a client, neither can the assistant working on their behalf.

Desktop AI actions are also tied to an enrolled device, and sensitive actions need approval.

FediPilot and other approved Fedix assistants can operate MyLedger through the same actions as the Copilot. That matters for audit: there is one path to review, not a separate back door for each tool.

The Copilot can work across reconciliation, reports, data, compliance, communication, e-signature, Xero, checklists and memory, and undo, finding the right action as it goes.

A read-only library of vetted Fedix workflows that AI assistants can follow.

If something goes wrong, you need to know who did what. Sensitive actions are recorded centrally, with categories and retention, as evidence of who did what. And if you need to step back, you can preview a plan to roll a client's ledger back to an earlier time, apply it, and undo it if needed.

Return preparation is where accountants feel the pull of automation most. In Fedix, an AI assistant works through the client's documents and produces a first draft of the return preparation for an accountant to review. The draft is the starting point; the accountant's review is the point. Around LodgeiT there is also a data helper: you can bring a legacy client book in from CSV, and export the list of returns awaiting lodgement in a format LodgeiT can take in.

Whatever product you evaluate, ask:

This is general information about software design, not advice about your own obligations.

The security overview is at https://home.fedix.ai/security?utm_source=devto&utm_medium=article&utm_campaign=cold-start-2026-10&utm_content=pub-3ae03af2820a, and the product pages start at https://home.fedix.ai/?utm_source=devto&utm_medium=article&utm_campaign=cold-start-2026-10&utm_content=pub-3ae03af2820a.

── more in #ai-agents 4 stories · sorted by recency
── more on @fedix 3 stories trending now
sponsored brought to you by zahid.host 4,200+ EU-deployed projects
reading about agents? ship yours in a single git push.

Run your AI side-project on zahid.host

EU-based hosting, git-push deploys, automatic HTTPS, no cold starts. Free tier with a custom domain — perfect for shipping the agent you just read about.

$git push zahid main
→ Live at https://your-agent.zahid.host ✓
Get free account → Pricing
from €0/mo · no card required
LIVE [news/letting-ai-near-the-…] indexed:0 read:3min 2026-11-22 · —