AI Risk Clock
Doomsday Clock2 min to midnight
→
🌑 Dark edit
itnews.com.au2026-05-28

Gov urges agencies to fix security basics before buying into frontier AI - iTnews

PolicySafety

Oh, brilliant. The Australian government's Department of Home Affairs has issued a PSPF advisory that basically says: 'Stop drooling over frontier AI and patch your bleeding servers.' In a stunning display of bureaucratic common sense — rare as hen's teeth — the advisory states that government entities do not need access to frontier AI models for effective cyber defenses, and should instead fix security fundamentals. Because apparently, the message that 'you can't bolt AI onto a compromised network' required an official government memo.

Consider the state of play: public sector IT systems are held together with sticky tape and prayers, yet every agency is clamouring for the latest LLM-powered threat detection. The Home Affairs advisory is essentially the adult in the room saying, 'Have you tried turning it off and on again?' before spending millions on a model that can generate incident reports in iambic pentameter. The PSPF — Protective Security Policy Framework — now includes this gem, presumably because too many procurement officers were asking for budget to buy GPT-10 while their Active Directory was held together by a password of 'admin123'.

The cruel joke, of course, is that this advisory will be read with nodding agreement and then promptly ignored. Agencies will continue to chase the frontier AI mirage because it sounds impressive in ministerial briefings, while the fundamentals remain as gaping as a SQL injection on a public-facing portal. When the inevitable breach happens, the review will note that 'despite clear guidance,' the agency had deployed an AI system on a network that hadn't been updated since the Howard years. But don't worry — the AI will write a very apologetic incident report.

Read this story in another voice
● REC · 2026