ASD Says The Fix Is Not In The Model. It Is In Who Holds The Laptop.

Device security for remote offshore employees is not a product you buy. It is a question of who owns the machine, who controls the account, and who can revoke both by lunchtime. The Australian Signals Directorate has just put that question at the centre of its AI guidance, because the technical fix everyone waited for is not coming.

Most Australian founders do not own that control layer. They rent it, informally, from a contractor. Outstaffer sells the alternative as managed equipment, so read this as a disclosed interest.

ASD moved the control point outside the model

ASD's September guidance on agentic AI systems, published on 11 September and reported by iTnews on Monday, finds that no fully reliable technical mitigation currently exists for prompt injection, because a language model cannot dependably tell instructions apart from information when both arrive in the same context window.

So ASD pushed the controls outward, into the software layer wrapped around the model. Least privilege on what an agent can reach. Human approval before high impact actions. Logging of prompts, tool calls and configuration changes. A permission system somebody administers.

Read that list as an employer, not a security engineer. Every item assumes you control the identity, the device and the access.

Can you revoke anything on a machine you do not own?

No. If your Manila developer works on their own laptop, under their own accounts, with their own AI subscriptions, your enforcement is an email asking nicely.

Device security for remote staff is whatever the engagement says it is

The contractor model fails here. Virtual Coworker and most of the Australian field treat the home office as the worker's problem, and for a decade that was a fair trade. The work was admin. The exposure was a spreadsheet.

That has stopped being true. The same person now runs an AI assistant wired into your inbox, your CRM and your client files, on hardware you have never seen, and the model itself will not protect you.

There is a classification edge, and it cuts the same way. Specify the device, the hours and the security posture and you are describing employment, not a contract for services: the five-question test Fair Work actually applies. Get advice on your own arrangement first.

The control layer costs about $11.90 a person per working day

Outstaffer publishes Employer of Record at $250 a month per employee. Five people is $1,250 a month, $15,000 a year, or roughly $11.90 per person per working day.

What does $250 a month actually buy?

A device you can remotely lock and wipe, data destruction to NIST SP 800-88 r1 before a departing employee's machine is collected, and management on a SOC 2 compliant platform. Underneath all of it, an employment relationship with an account you issued and can close. Somebody still has to pay for the hardware, and that number moved this year.

The contractor version costs nothing a month and delivers none of it. Zero is a real saving right up to the morning you need to switch someone off.

The strongest argument against me is the base rate

Thousands of these arrangements have run for years without incident, and a contractor's laptop causing a reportable breach is rare today. A managed device does not stop someone photographing a client list either, and ASD wrote for organisations with a security function, not a 22 person agency in Richmond.

All true. None of it changes the direction: the tooling your team uses got far more capable this year while the boundary around it did not move.

Related reading

• Wages Up, Output Per Hour Flat: What The June Quarter Actually Cost You
• Fair Work Will Not Hear Your Remote Dismissal. Another Regulator Will.

Device security is decided when you choose how to engage someone

ASD has said the model is not the control. For a small Australian business the control is not software, it is an arrangement: whose laptop, whose account, whose off-switch. Device security for remote offshore employees gets settled the day you decide how to engage the person, and almost nobody treats that as a security decision.

If you want the boundary rather than the invoice, run one role through it properly.

So, the real question. If your best hire in Manila vanished tonight, how long would it take you to lock their machine, and who would you have to ask?