Hundreds of conversations with an AI chatbot have been found sitting out in the open on the public web. The BBC reports that chats with Anthropic’s Claude were discovered publicly accessible – not through a break-in, but because of the way conversations had been shared.
Separately, security researchers showed that Claude Cowork, an AI tool for the Mac, could escape the sandbox meant to limit it and reach files across the machine.
Neither story is a reason to ban AI at work. Both are a reason to be deliberate about it.
Why this matters for your business
Staff paste real work into AI tools every day: client emails, quotes, contracts, spreadsheets, and occasionally passwords. If a conversation is shared using a link, that link can be found by other people. If an AI tool has broad access to a computer, it can read whatever that computer can read.
Four practical steps
- Write a one-page AI policy. Say plainly what may and may not be pasted into a chatbot. Client data, payroll and passwords belong on the “no” list.
- Use business accounts, not personal ones. Business tiers generally keep your data out of model training and give you admin oversight of who is using what.
- Treat every “share” link as public. If you share a conversation by link, assume it can be found.
- Limit what AI tools can reach. Give them the folders they actually need, not the whole machine.
Most small businesses we speak to have not had the AI conversation yet. Setting sensible rules takes about an hour, and it is a great deal cheaper than explaining a leak to a client.
Worried this affects your business? Get a free 15-minute IT check – call Trends IT on 0485 011 911 or visit /contact/.