AI governance for Australian schools.
External Signals is a monthly briefing on what's shifting outside Australian independent schools — regulation, technology, sector incidents, funding, and social change — and what it means for the people running them. Produced by Arvoe for principals, business managers, and board members across the sector.
Top 5 signals most relevant this month
Page 4- 01
eSafety warns schools their published photos are being used to make AI fakes
AI & Data Governance - 02
Commonwealth creates an Office of AI and names schools as a workstream
AI & Data Governance - 03
Two Queensland school stabbings in two days; state rules out principal search powers
Safety & Wellbeing - 04
First national data on AI in child sexual image abuse: teenagers disclose to chatbots first
Safety & Wellbeing - 05
Victoria rebuilds the Working with Children Check
Regulatory & Compliance
Can you name every person acting for your school in a child-facing role, including those you do not employ, and say what screening each of them holds? Six signals across four jurisdictions widened that boundary this month, and Victoria's new liability rule also covers past conduct.
Quieter than June. Financial & Operations runs warmest, on a single liability change that covers past conduct. AI & Data Governance averages cool but carries the month's highest-scoring signal.
At a glance
Two things ran through July. Most of the month's compliance weight came from rules already passed: a Victorian liability change that covers past conduct, a reportable conduct scheme that changed regulator in February, and NSW reporting definitions in force since August 2025. Nothing was announced, and the obligations applied anyway.
The other was AI. The Prime Minister created an Office of AI inside his own department and named AI in schools as one of five ministerial workstreams. Two weeks later eSafety told schools that AI fakes of staff and students are being made from photos the schools published themselves.
Bars show how many signals touch each area; one signal can touch several, so they total more than the verified count. The dot shows how hot the area ran.
When each signal is expected to bite. The hero figure is the first two buckets combined.
National: National Cabinet takes up the proposed Australian Standards for AI, date not yet announced. The first test of whether the states sign up, and the point at which any schools-specific position would surface.
UK: Keeping Children Safe in Education 2026 in force from 1 September, with statutory RSHE deepfake content in use from the same term. The closest settled answer to this month's AI imagery signals, though not an Australian one.
NSW: the last two Children's Guardian Amendment Act provisions commence. Private health facilities serving children become reporting entities, and "substitute residential care" loses its two-nights-in-seven threshold.
ACT: the domestic and family violence information-sharing scheme commences, with schools designated as information sharing entities.
EU AI Act: the new prohibition on AI-generated non-consensual intimate imagery and child sexual abuse material applies. No Australian obligation, carried as the offshore comparator to this month's top signal.
National: the automated decision-making transparency obligation under the Australian Privacy Principles commences. The one AI obligation that already has a firm date.
NSW: a compliant anti-bullying policy becomes a condition of school registration, with NESA spot checks.
National: education ministers have agreed that sectors will embed the National Framework, or an equivalent jurisdictional one, and that schools will publish or link to aligned policies.
Risk weather this month
July was quieter than June. Four signals scored 7 or above, against nine last month, across 23 signals rather than 32. The highest-scoring signal is not a compliance date: it is a regulator saying harm is already happening. Most of the month's obligations started earlier and only surfaced now.
The ratings below combine severity with timing, not volume, so a small area can rate higher than a busy one.
| Risk area | What is driving it | This month | Vol. |
|---|---|---|---|
| Financial & Operations | Victoria widened who a school can be held liable for, to include volunteers and religious ministers. It covers conduct before commencement, so the question is insurance response rather than policy. | WARM | 1 |
| Safety & Wellbeing | Two Queensland school stabbings in two days, the first national data on AI-generated sexual imagery of Australian teenagers, a parent-driven boarding discipline complaint, and UK guidance bringing AI-generated imagery into statutory safeguarding. | MILD | 4 |
| Regulatory & Compliance | Instruments already passed reaching commencement across Victoria, NSW and the ACT, including one NSW change in force since August 2025. Little of it is new law; most of it is new obligation. | MILD | 5 |
| Workforce & IR | Victorian government-school bargaining stayed deadlocked through a statewide strike and continuing bans. Separately, new Australian research on sexism and abuse directed at teachers, disproportionately at women. | MILD | 3 |
| Cyber & Data Security | A national advisory on credential exposure affecting Fortinet firewalls and VPN gateways, with steps to apply now. Alongside it, a national education department breached through a third-party portal. | MILD | 2 |
| AI & Data Governance | Mostly early-stage policy on long horizons. At the other end sit the month's two highest-scoring signals: a regulator telling schools their published imagery is being harvested and manipulated now, and a new federal AI office. | COOL | 4 |
| Teaching & Curriculum | Classroom material, none of it binding here and most already in effect: a federal social cohesion hub, an international AI literacy framework, a Norwegian ban on generative AI in the primary years, and a UK addition to statutory relationships education. | COOL | 4 |
Each area's rating combines how serious its signals are with how soon they land. Warm means worth watching now; mild means real but not pressing.
The count of signals in the area this month; areas sum to the month's 23. It is separate from the rating: an area can carry a real volume of work while still reading mild.
Top 5 signals most relevant this month
eSafety warns schools their published photos are being used to make AI fakes
On 28 July the eSafety Commissioner published an Online Safety Advisory for Australian schools. Between January and March 2026 eSafety received more than 100 reports of anonymous accounts targeting schools and staff: AI-generated videos, face swaps, memes and fabricated stories about principals and teachers. Almost all used images taken from school websites and social media accounts. Staff headshots were used to make fake videos and teacher rating posts. Most of the activity sat on anonymous TikTok and Instagram accounts.
The Commissioner also said eSafety knows of a significant number of recent cases involving AI-generated child sexual exploitation material in school settings, without attaching a number. The Advisory asks schools to take a risk-based approach to what they publish, secure clear consent, check who can access shared content, review privacy and security settings, and hold a response plan. eSafety separately reports seven "nudify" services withdrawn from the Australian market or brought into compliance.
The exposure comes from enrolment marketing, which usually sits outside child-safety governance. Open day photos, the musical, the First XI, staff profile pages: published deliberately, at high resolution, with names attached, and nobody involved is breaching school policy. eSafety also notes that not all of this material is removable under the Online Safety Act 2021, particularly content targeting adult staff, so a teacher whose headshot becomes a fake video may have no takedown path. Where there is no remedy, the only control is what the school chose to publish.
Worth checking — How many identifiable, high-resolution photos of named students and staff are on our website and social accounts now, and who approved them? If a fake video of our principal circulated tonight, who do we call first, and what could actually be removed? Does our consent process cover how a staff member's image may be published?
Commonwealth creates an Office of AI and names schools as a workstream
On 15 July the Prime Minister announced an Office of AI inside the Department of the Prime Minister and Cabinet, giving Australian AI policy a single coordinating owner for the first time. He also announced a set of Australian Standards for AI, describing the framework for large AI data centres as "clear, consistent and mandatory". That framework covers where data centres are built, an obligation to be net generators of energy, and their water use, alongside copyright and artist protections. It does not extend to organisations that merely use AI, and creates no obligation for any school.
Setting out the workstreams the new office will coordinate, he named five: energy, copyright, productivity, employment, and education. He noted the Education Minister was meeting his counterparts that day on the impact of AI in schools. The Prime Minister will seek agreement from premiers and chief ministers at a National Cabinet this month, and the government "will aim to bring the legislation to Parliament early next year". No mandatory obligation has a date yet.
There is no new obligation for schools here. What changed is the policy position. For two years a board could reasonably treat AI governance as unsettled and wait. The Commonwealth has now created a coordinating office, used the word mandatory about the first framework it intends to build, and listed AI in schools among five workstreams. Two limits are worth being precise about: schooling is a state and territory responsibility, so the federal lever is coordination rather than standard-setting for schools; and the automated decision-making transparency obligation commencing 10 December 2026 is a separate instrument that this announcement does not change.
Arvoe builds AI governance software for Australian schools. Closing this gap is what we do: a documented position on how AI is used across the school, with the controls and evidence a board can see behind it. Click the link below to book a conversation about where your school stands.
Book a conversationWorth checking — When National Cabinet takes this up, who reads the communiqué, and what would we do if education were named a covered domain? If a parent or board member asked today what our AI governance position is, could we hand them a document? What is the gap between what staff already do with AI tools and what our policy permits, and when did anyone last measure it?
Two Queensland school stabbings in two days; state rules out principal search powers
Queensland recorded two high-school stabbings on consecutive days. On Monday 13 July a 15-year-old was stabbed at a state high school with abdominal wounds and remained stable; a 15-year-old was charged with acts intended to cause grievous bodily harm and refused bail. On Tuesday 14 July a 17-year-old was stabbed multiple times at an independent religious school with life-threatening injuries and was later stable in intensive care; a 16-year-old was charged with acts intended to maim, disfigure or disable and with unlawful possession of a weapon, and was granted bail. Charge details are police allegations reported by credible press.
Queensland's Police Minister then ruled out giving principals handheld metal-detector search powers, saying the government did not want to make this principals' job, and pointed to existing police powers under Jack's Law, which allow police to scan with handheld detectors in any public place without a warrant.
Weapons response in Queensland stays police-led. A school's own capability is its behaviour policy, its escalation thresholds and its relationship with local police, not a power to search. Two incidents within 24 hours, both in school hours, is also the pattern that drives copycat risk and parent anxiety, so the position may not hold.
Worth checking — Does our critical-incident plan route a weapon report to police rather than a staff search, and does every deputy know that? Who decides to lock down, and would that decision be made the same way at 3pm Friday as 9am Monday? If an incident elsewhere reached our families through social media first, what would we send in the first hour and who signs it?
First national data on AI in child sexual image abuse: teenagers disclose to chatbots first
The Australian Cybercrime Observatory at Adelaide University, with ICMEC Australia and the AFP-led Australian Centre to Counter Child Exploitation, released Australia's first nationally representative study on AI in online child sexual victimisation. It surveyed 1,894 Australians aged 16 to 18 in early 2026.
At least one in 25 has experienced, or has a close friend who has experienced, online sexual victimisation involving AI before the age of 18. AI was involved in 26 per cent of non-consensual sexual image cases. Boys were disproportionately targeted where AI was involved: 26.6 per cent of boys' victimisation against 9 per cent of girls'. On disclosure, 18.7 per cent of victims told an AI chatbot, against 13.2 per cent who told a teacher, doctor, counsellor, police officer or helpline. The full report was not publicly posted at the time of writing; the figures come from the collaboration's media release and were corroborated across independent outlets.
Child safety obligations assume the school finds out. Reportable conduct schemes, child safe standards and the escalation policies behind them are all triggered by someone telling someone. The disclosure finding says students are more likely to tell an app than any trained adult in their life. The obligation is unchanged; the chance of it being triggered is lower. Separately, an AI-generated image of a student is the same safeguarding matter as a photograph of that student, and an image-based abuse response written before generative tools were common may not read that way.
Worth checking — Where in our year do we actually tell students what to do when an image of them is circulating, rather than waiting for them to ask? If a student uses a school-provided AI tool to disclose distress, is that logged, escalated or lost? Could we say what happens in the first hour after a disclosure, and does the answer change if the image was generated rather than taken?
Victoria rebuilds the Working with Children Check
From 28 July 2026 Victoria changed how the Working with Children Check operates, under the Social Services Amendment (Child Safety, Complaints and Worker Regulation) Act 2025, responding to the Rapid Child Safety Review. Three things changed.
Internal review by the Social Services Regulator replaced VCAT merits review for anyone refused a clearance or whose clearance is revoked; applications must be made in writing within 28 days and are decided on the written material, with no hearing. Interim bar and suspension decisions are not internally reviewable at all. The scheme can now consider a wider range of child-safety risk information, expressly including matters that did not result in a charge, conviction or finding of guilt, and information from other states and territories. New interim bar and suspension powers, effective the same day, let the regulator stop a person working with children while it assesses risk.
A current clearance is now more contingent than it was. It can be suspended mid-employment, on a wider evidence base than a criminal record, with no tribunal to appeal to. The practical question is no longer whether every worker holds a clearance, but what happens in the hours after one is barred. Further reforms are flagged for later in 2026, including mandatory training for applicants and a system requiring employers to verify worker and volunteer status.
Worth checking — If a staff member or volunteer were interim-barred tomorrow, who would find out, how quickly, and what would we do that day? How often do we re-verify clearance status for people already working with us, rather than at hire? Have we told staff that tribunal review is gone, before someone needs it?
For your next board meeting
We read the month's signals as evidence of governance questions the board already owns. One matter this month.
Who is acting for your school, and would your screening hold?
Every board owns the question of who acts for the school in a child-facing role: employees, but also volunteers, contractors, chaplains, coaches, homestay hosts, the parent who drives the minibus. That position is usually set out in a screening policy and a reporting policy, and reviewed when something forces a review. The question is not new. What keeps moving is where the boundary sits.
Six signals across four jurisdictions changed who a school is responsible for, what it must report, or what evidence can be used to keep someone out. Taken jurisdiction by jurisdiction:
- All jurisdictions have agreed to make screening national. In November 2025, through the Standing Council of Attorneys-General, every state and territory endorsed the Agreement to Deliver National Working with Children Check Reform, setting six priority actions toward more consistent working with children check and blue card systems. These include mutual recognition of negative notices, so a refusal in one jurisdiction is visible in another. There is no single commencement date; each jurisdiction must legislate.
- Victoria made three changes at once. The Justice Legislation Amendment (Vicarious Liability for Child Abuse) Act 2026 extends liability to child abuse by people in roles akin to employment, expressly including religious ministers and volunteers but not independent contractors, and it reaches conduct from before commencement as well as after. The Working with Children Check was rebuilt on 28 July, with a wider evidence base, new interim bar and suspension powers, and internal review replacing tribunal review. Reportable conduct and the child safe standards also moved to the Social Services Regulator, so notifications now go there.
- NSW widened what must be notified, a year ago. Changes to its reportable conduct scheme definitions commenced on assent in August 2025, not on the October 2026 date the remaining two schedule items carry.
- The ACT will require schools to share information. From 24 November 2026, subject to commencement, schools become information sharing entities under the territory's domestic and family violence scheme, and will in some circumstances have to share risk information with other entities.
- The reporting pathway has not widened with them. In the national study above, 18.7 per cent of victims aged 16 to 18 told an AI chatbot, against 13.2 per cent who told a teacher, doctor, counsellor, police officer or helpline.
Can we produce today a list of every person acting for this school in a child-facing role, including those we do not employ, with the screening each of them holds and who is accountable for checking it?
The exposure is not from any single change in July. It is that the definition of who counts keeps widening, and that liability now reaches back as well as forward. Victoria's retrospective provision makes that concrete: exposure attaches to arrangements made years ago, under policies nobody currently in the room wrote. Screening records that were an administrative matter become evidence.
A first paper does not need a legal opinion. It needs four things: a register of every category of person with child-facing access, employed or not, with the screening each holds and a named owner; a statement of which jurisdictions the school operates in and which reportable conduct scheme applies to each; what happens operationally in the first 24 hours after a clearance is barred or suspended; and what the school actively tells students about reporting, as distinct from what it makes available if they look. Where volunteers, chaplains or coaches come through a third party, say who holds the screening records and whether the school can obtain them. That is usually the gap.
Received this as part of a board pack? The monthly briefing is free to school leaders at arvoe.ai/external-signals.
Signals by risk theme
Regulatory & Compliance
5 signals| Signal | Date | State | Source | Score |
|---|---|---|---|---|
| NSW widens reportable conduct definitions, in force since Aug 2025; two further items follow Oct 2026 | 15 Aug 2025 | NSW | Children's Guardian Amendment Act 2025PRIMARY | 6 |
| Victoria overhauls the Working with Children Check | 28 Jul 2026 | VIC | Victorian GovernmentPRIMARY | 6 |
| Victoria's reportable conduct scheme and child safe standards move to the Social Services Regulator | 23 Feb 2026 | VIC | Commission for Children and Young PeoplePRIMARY | 5 |
| ACT designates schools as information sharing entities under its domestic and family violence scheme | 24 Nov 2026 | ACT | School GovernanceCREDIBLE | 4 |
| All jurisdictions endorse the Agreement to Deliver National Working with Children Check Reform | 14 Nov 2025 | AU | National Office for Child SafetyAUTHORITATIVE | 3 |
Nothing here was decided in July, and only one of the five started in it. The rest have been in force since February or earlier, or are still ahead. That is what makes the group easy to miss: there was no announcement to react to, and the obligations applied anyway. More people are in scope, more information can be used against an application, and more of what happens inside a school must be reported outside it.
Safety & Wellbeing
4 signals| Signal | Date | State | Source | Score |
|---|---|---|---|---|
| Two Queensland high-school stabbings in two days; state rules out principal search powers | 15 Jul 2026 | QLD | Australian Community MediaCREDIBLE | 7 |
| First national study: AI involved in 26% of non-consensual sexual-image cases among 16 to 18 year olds | 7 Jul 2026 | AU | Australian Cybercrime ObservatoryCREDIBLE | 7 |
| Parents allege psychological harm over boarding discipline at an independent school in Victoria | 30 Jun 2026 | VIC | The ConversationCREDIBLE | 5 |
| UK publishes Keeping Children Safe in Education 2026, bringing AI and deepfake harms into statutory safeguarding | 7 Jul 2026 | UK | UK Department for EducationAUTHORITATIVE | 3 |
Two of the month's four highest-scoring signals are here, and they describe harm arriving two different ways: one physical and visible, one digital and largely invisible to the adults in the building. The boarding matter is different again. No regulator has been reported as intervening, and the school moved from an internal decision to national coverage within weeks on parent complaint alone. The UK guidance is the only settled answer here, and it is not an Australian one.
Teaching & Curriculum
4 signals| Signal | Date | State | Source | Score |
|---|---|---|---|---|
| Federal Social Cohesion Education Hub launched, with a curriculum review attached | 30 Jun 2026 | AU | Ministers' Media CentrePRIMARY | 5 |
| UK adds AI and deepfake content to statutory relationships, sex and health education | 16 Mar 2026 | UK | UK Department for EducationAUTHORITATIVE | 3 |
| OECD, European Commission and Code.org publish an AI Literacy Framework for primary and secondary education | 17 Jun 2026 | GLOBAL | AILit FrameworkAUTHORITATIVE | 3 |
| Norway bans generative AI during the school day for grades 1 to 7, and allows it under supervision for ages 14 to 16 | 19 Jun 2026 | GLOBAL | The Next WebCREDIBLE | 3 |
Three of these four are decisions about what to teach children about AI, or whether to let them use it, and they point in different directions. Norway restricts it in the primary years. The OECD framework treats AI literacy as a competency to build from primary onwards. The UK adds deepfakes to statutory health education. No Australian position has been set. The federal hub is the only domestic item and is not about technology: about half its resources address antisemitism, as does the curriculum review attached to it.
Workforce & IR
3 signals| Signal | Date | State | Source | Score |
|---|---|---|---|---|
| AEU Victoria resumes protected industrial action in government schools and restarts bargaining | 13 Jul 2026 | VIC | EducationHQCREDIBLE | 5 |
| AEU Victoria holds a 24-hour statewide strike across government schools | 23 Jul 2026 | VIC | Human Resources DirectorCREDIBLE | 5 |
| Adelaide study documents worsening sexism and misogyny directed at teachers | 26 May 2026 | AU | Adelaide UniversityAUTHORITATIVE | 5 |
The Victorian dispute creates no obligation for non-government schools, but sets a benchmark they will be measured against. Members rejected an offer in the high twenties to low thirties across four years, citing workload rather than pay alone. The Adelaide study is the same picture from another angle: sexist and abusive student conduct directed mostly at women, which most schools treat as a pastoral matter rather than a workplace hazard with a duty attached.
AI & Data Governance
4 signals| Signal | Date | State | Source | Score |
|---|---|---|---|---|
| eSafety Advisory: AI manipulation of school-published imagery, 100-plus reports of accounts targeting schools and staff | 28 Jul 2026 | AU | eSafety CommissionerPRIMARY | 8 |
| Federal Office of AI established in PM&C, Australian Standards for AI announced, AI in schools named a coordinated ministerial workstream | 15 Jul 2026 | AU | Prime Minister of AustraliaPRIMARY | 7 |
| EU postpones AI Act high-risk deadlines and adds a prohibition on AI nudifiers and child sexual abuse material, applying 2 Dec 2026 | 27 Jul 2026 | EU | Official Journal of the EUPRIMARY | 3 |
| Anthropic launches Claude for Teachers, a free assistant for verified educators, United States only | 14 Jul 2026 | US | ChalkbeatCREDIBLE | 3 |
Australia spent July building the machinery to answer questions it has not yet answered, while its own regulator said the harm is already happening. One signal is a framework with no obligations attached and a National Cabinet discussion still to come. The other is a regulator reporting more than a hundred cases in a single quarter and asking schools to change what they publish this term. The EU moved both ways in one instrument: high-risk compliance, which covers education uses such as admissions and exam proctoring, deferred to 2027 and 2028, while a new prohibition on AI-generated intimate imagery and child sexual abuse material applies from 2 December 2026. Claude for Teachers is US-only with no Australian release, included because the pace of the classroom-AI market is itself the signal.
Cyber & Data Security
2 signals| Signal | Date | State | Source | Score |
|---|---|---|---|---|
| ACSC advisory on a widespread credential-exposure campaign against Fortinet firewalls and VPN gateways | 18 Jun 2026 | AU | ASD's Australian Cyber Security CentrePRIMARY | 5 |
| UK Department for Education breached, 607,000 contact records taken via a scheme portal and its online help desk | 30 Jul 2026 | UK | BBC NewsCREDIBLE | 3 |
The advisory is the clearest item in the edition: it names the vendor, the exposure and what to do about it. The difficulty is ownership. In most schools this equipment is managed by someone who does not report to the person reading this. The breach is the same lesson from the other end: 607,000 contact records lost through a scheme portal and a help desk, neither of which anyone would have named as critical.
Financial & Operations
1 signals| Signal | Date | State | Source | Score |
|---|---|---|---|---|
| Victoria extends vicarious liability for child abuse to individuals akin to employees, retrospectively | 24 Feb 2026 | VIC | Parliament of VictoriaPRIMARY | 6 |
One signal, placed here rather than under regulation because the exposure is financial before it is procedural. The Act reaches people in roles akin to employment, including volunteers and religious ministers, and excludes independent contractors. It applies to abuse before as well as after commencement, and allows some settlements to be set aside. The question is not whether the school's child-safety policy is current, but whether its abuse-liability cover responds to claims involving people it never employed.
How this briefing is made
Every month we scan the full policy, technology, social, economic, legal and environmental landscape affecting Australian schools, across all sectors rather than only independent. From that, we surface only what clears verification. Fewer, higher-confidence signals beat a large scan count. This month, 23 signals cleared the bar. Signals are included by the date they cleared verification; the event date shown may fall either side of the period.
Each signal is scored for relevance to schools and placed on a horizon (how soon it lands) and a set of risk domains. The risk weather combines severity and immediacy into a single read per area, so you can see where the pressure sits and track it month to month.
Every signal is checked against source, and load-bearing claims are read off the primary where one exists. Every source is shown with its authority, so you can see what a claim rests on. We do not name schools involved in adverse incidents. The lesson travels without the name.
This is a briefing to help you see what is coming and ask better questions. It is not compliance advice, and not a substitute for your own legal, IR or child-safety processes.
Want a signal read against your school's specific profile?
Arvoe helps schools turn signals like these into a governed, board-ready position. If the board question above raised something the school cannot yet answer with confidence, our quick interactive assessment is a free place to start. To have a signal in this edition read against your school's specific profile, reply to the email this briefing arrived with.
We help schools unlock AI with confidence, so our kids can thrive in an AI world.
Ryan Speak · ryan@arvoe.ai · arvoe.ai/external-signals