In the World of AI: Week of 18 August 2026, Xero Puts AI Into Bookkeeping and Claude Remembers You

Xero has announced a batch of AI bookkeeping tools, a new report says scams against small firms are fewer but costlier, and Claude now carries your context from one conversation to the next. Here is the week to 25 August 2026, with one thing to do about each story.

1. Xero announces AI tools for small businesses and accountants

Xero's media release of 20 August, datelined 19 August and issued at Xerocon US, lists a set of AI features. They include a Partner Hub that shows book health and month-end readiness across a practice's clients, and checks that flag unreconciled items, duplicates, missing documents and anomalies. Smart Document Capture pulls data from source documents into Xero, and Auto Bank Reconciliation matches high-confidence transactions automatically. Document Requests lets the assistant ask clients for documents, send reminders and match them to transactions, with customer approval at each step. Xero also announced Cash Flow actions, Payment follow-ups and Bill Protection, plus integrations that bring live Xero data into Microsoft 365, Claude and ChatGPT. A month-end agent called XeroForce is in early access, with general availability "later this year".

The release does not say when or whether these features reach Australia. It limits the new payments capabilities and Xero Payroll, powered by Gusto, to US customers.

Why it matters: Bookkeeping is repetitive, which makes it a natural place for AI. The release keeps a customer approval step in the document requests, which is the kind of human check worth asking about in any tool.

This week: Ask your bookkeeper or accountant which of these features are live for Australian customers, and which task they would hand to AI first. Our answer on what AI can automate in a small business helps with that shortlist.

2. Small business scams are fewer but cost more

SecurityBrief Australia reported on 25 August on a BizCover analysis of Scamwatch reports from 2024 and 2025. Small business scam reports fell 20.8%, total losses rose 3.4% and the average loss per report rose 30.4%. BizCover's Akshaye Kalkura said criminals now aim at single high-value payments, such as changed supplier bank details or fake invoices that slip past rushed checks. Attackers impersonate suppliers, executives or customers using accurate business details.

The article also quotes an identity specialist who says spotting spelling errors or odd language is no longer a reliable test, because of AI-generated content and deepfakes. Kalkura recommends verifying unexpected payment or bank-detail requests through a separate channel before acting.

Why it matters: Fewer reports with a larger average loss means a single missed check does more damage. The scammers in the article rely on rushed checks.

This week: Require a second person, or a phone call to a known number, before anyone changes supplier bank details or pays an unusual invoice. Write the rule down so a new starter follows it too.

3. Claude gets shared memory across chat and Cowork

Anthropic announced on 25 August that Claude's memory is now shared between chat and Claude Cowork, and updates during conversations instead of afterwards. It is on by default for Free, Pro, Max and Ultra plans. Users can see what Claude remembers under Settings, read, edit or delete each item, pause memory or reset it. Sensitive topics such as health or religious beliefs are not saved by default, and Claude does not store sensitive ID numbers, criminal history or immigration status even when the sensitive-topics option is on. On Team and Enterprise plans, administrators control whether memory is available.

Why it matters: A tool that remembers client names, preferences and past work saves retyping. It also keeps a record of what people have told it, which matters if staff use personal accounts for work.

This week: If your team uses Claude, decide whether memory should be on for work accounts, and tell staff what not to put into it. Our answer on whether the Privacy Act applies to AI tools covers the privacy side, and you should take advice for your own situation.

4. OpenAI cuts GPT-5.6 Sol pricing for three months

OpenAI announced on 21 August a price reduction of over 20% on GPT-5.6 Sol for API use and for Codex credits and ChatGPT Work credits on eligible plans. In the API, input falls from US$5.00 to US$4.00 per million tokens and output from US$30.00 to US$20.00. The cut is described as running for three months. Pro, Plus and Business subscription usage remains unchanged.

Why it matters: The cut reaches businesses that pay per use through a custom tool, not those on a flat subscription. A temporary price is a poor basis for a budget.

This week: If someone built a tool for you on OpenAI's API, ask whether the running cost has dropped, and whether your quote assumed the temporary rate.

What to do this week

• Ask your bookkeeper which Xero AI features are available in Australia.
• Add a second-person check before any change of supplier bank details.
• Decide whether Claude memory is on for work accounts, and tell staff what to keep out of it.
• Ask whoever runs your API-based tools how the temporary Sol pricing affects your costs.

If you would like help working out where AI fits in your business, see our AI adoption service.

Sources

• Xero, "Xero Announces New AI Innovations for Small Businesses and Accountants at Xerocon US", 20 August 2026: xero.com
• SecurityBrief Australia, "AI fuels targeted scams against Australian small firms", 25 August 2026: securitybrief.com.au
• Anthropic, "Claude's memory works everywhere, and you decide what's in it", 25 August 2026: claude.com
• OpenAI Developer Community, "20% price reduction for GPT-5.6 Sol: API, Codex credits and ChatGPT Work", 21 August 2026: community.openai.com