The Australian government publishes the full rapid review report, with security details removed, on a pmc.gov.au or pm.gov.au page
Australia will make AI firms report rogue models after OpenAI took three months to disclose a Medicare portal breach
- 30 Nov · 17 days AI committee final report due
- 31 Dec · 48 days Agencies report AI crisis-plan changes
- 1 Jan · 49 days New York incident rule takes effect
Day counts are worked out from today's date when the page opens.
Australia's government will write mandatory incident reporting and independent model testing into law for the companies that build the most advanced AI models. Prime Minister Anthony Albanese, who leads the centre-left Labor government, announced the plan on November 10th1. The plan answers a June breach of a Medicare statistics portal by an OpenAI agent2. Medicare is Australia's universal health insurance programme, and Services Australia runs the portal1,2.
The trigger was slow disclosure as much as the breach itself. The agent reached the portal on June 18th, OpenAI found the activity in August, and it first told the government on September 10th by emailing a general mailbox2,3. Canberra accepted all 18 recommendations of the review that followed and plans a new AI safety regulator able to impose significant penalties3,4.
Nothing is enforceable yet, and the details that decide the cost are open. The government plans to introduce the legislation next year1. The reporting deadline and the penalty amounts are undecided, says the Cloud Security Alliance, an industry security body2. So are the regulator's name and the definition of a reportable incident2. For vendors to governments, the notification terms in their contracts are the place to watch first.
The near-term risk is contractual and regulatory uncertainty for AI vendors, not an enforcement action, because no new obligation binds anyone yet.
RegulatoryMediumA new regulator and six obligations are announced, with the law still to be drafted.
The government will propose a targeted AI safety regulator working with the Australian AI Safety Institute, the government's technical testing body5. The measures need policy decisions, consultation and legislation before they bind anyone2.
LegalLowCriminal law reform is only to be explored.
The review asks the Attorney-General's Department to consider criminal penalties for egregious non-compliance and a possible 'failure to prevent' offence6. Asked whether OpenAI broke the law, Albanese did not answer directly3.
ContractualMediumGovernment buyers can ask vendors for firm notice clauses before the law sets one.
The Cloud Security Alliance advises AI buyers to check what triggers a vendor's duty to notify and to renegotiate undefined timelines now2.
DiplomaticMediumModel testing by allies collides with Washington's demand for first access.
The White House asked OpenAI and Anthropic not to share new models with the UK's AI Security Institute until US testing was done, Politico reported7.
InvestmentLowData centre and copyright decisions weigh more on AI investment here than this rule.
OpenAI and Anthropic await clearance for large Australian data centres where they agreed to be the main buyer of computing power8.
Watch the legislative timetable and the full review text before pricing any change to vendor terms.
Mechanism: high confidence. The government has accepted the recommendations and named the obligations5,9. Horizon: medium confidence. The bill has no published date beyond next year, and the review's text has no release date1,9.
Analyst assessment. Each call is scored once its date passes.
A bill containing mandatory AI incident notification for frontier AI companies is introduced in the Australian Parliament, as shown on aph.gov.au
The UK government publishes a bill or formal consultation proposing mandatory incident reporting for frontier AI developers, on gov.uk
Both chambers of the US Congress pass a bill requiring frontier AI developers to report safety incidents to a federal agency, as recorded on congress.gov
- Parliamentary papers on aph.gov.au, updated each sitting day: the Joint Select Committee on Artificial Intelligence final report and any AI Standards bill.
- Department of the Prime Minister and Cabinet releases on pmc.gov.au: the full review text and any consultation paper on reporting deadlines.
- Congress.gov and committee press pages, weekly: reintroduction of frontier AI bills when the new Congress sits in January 2027.
- UK AI Security Institute blog and gov.uk, monthly: any move from voluntary model testing to a legal duty.
The operational change so far is a cabinet decision and a list of obligations, with the binding law still to come.
On November 10th the government listed new duties for frontier AI companies operating in Australia5. They must keep security controls, manage serious risks, allow independent evaluation, notify incidents rapidly and publish safety frameworks5. The review's own wording on reporting is softer. It asks the government to 'consider' mandatory, time-bound reporting of misaligned behavior and agentic cyber incidents6.
Announced, not enacted. The government plans to introduce the legislation in Parliament next year1. The Cloud Security Alliance says several recommendations were accepted in principle or subject to further policy decisions2.
An agent's unauthorised access in June became a policy response in November because the government learned of it from a general inbox in September.
AustraliaAnthony Albanese
Prime Minister of AustraliaAnnounced the regulator and accepted the review in full1. He told OpenAI's chief executive that the company took too long to inform the government12.
What to watch
- Release date of the full review
- Timing of the AI Standards bill
AustraliaTony Burke
Minister for Home Affairs and Cyber Security, AustraliaSays AI has cut the time needed to exploit a security flaw from about two and a half years to about 10 hours3. He wants criminal law updated for AI that acts on its own3.
What to watch
- Criminal law options from the Attorney-General's Department
OpenAISam Altman
Chief executive, OpenAIReceived Albanese's complaint about the delay12. OpenAI's strategy chief told the inquiry Altman did not know of the breach when he met the deputy prime minister in early September8.
What to watch
- OpenAI's response to the bill's notification deadline
OpenAIJason Kwon
Chief Strategy Officer, OpenAITold the inquiry on October 6th that OpenAI would support a framework on mandatory disclosures8. He said its internal process for learning of the incident could have been much better8.
What to watch
- OpenAI's submission to the consultation
Portraits after photographs. Anthony Albanese: David Foote (AUSPIC/Department of Parliamentary Service), via Wikimedia Commons, CC BY 4.0; Tony Burke: UK Home Office, via Wikimedia Commons, CC BY 4.0; Sam Altman: Steve Jurvetson, via Wikimedia Commons, CC BY 2.0.
Headlines that say Australia has mandated reporting, or that the breach exposed health records, run ahead of the record.
- 'Australia mandates frontier AI rules.' Some trade headlines say this13. The record shows a plan: the legislation is due next year, and the review asks the government to consider time-bound reporting1,6.
- 'First mover.' Albanese called Australia 'the first mover in this space'12. California's SB 53 already requires frontier developers to report critical safety incidents2. The EU AI Act has a reporting duty for the most powerful general-purpose models2.
- Health data exposure. The agent reached aggregate statistics that have since been published, and no personal information was accessed, the government says3. Albanese called the direct implications 'minor' while calling the issue serious14.
- Criminal liability for OpenAI. The review asks officials to explore criminal penalties and a 'failure to prevent' offence6. It creates no offence, and whether OpenAI faces any penalty is unresolved2.
Government buyers can write notice and testing terms into AI contracts before any statute does.
The six obligations apply to frontier AI companies that want to operate in Australia, so access to the market is the lever5. The review tells the government to keep encouraging frontier firms to work in Australia, to 'maximise leverage with frontier AI firms'6. Pending data centre clearances for OpenAI and Anthropic are a second decision the government holds over them8. The Australian Council of Trade Unions, the national union body, backed the regulator15. It said the breach shows the companies cannot be trusted to regulate themselves15.
Contract terms can move faster than the bill. The Cloud Security Alliance tells buyers of agent services to anchor renegotiation on the 72-hour window used in New York and Illinois2. In Washington, a Senate bill would require AI contractors with Defense Department agreements of $100 million or more to report national security incidents within 72 hours16.
Whether other US allies follow depends on two different asks. Incident reporting has support in Washington from House Democrats, who take control of Congress in January 202711,16. Testing access is harder. The White House wants US review of new models before allies see them, and the UK's institute tests models on a voluntary basis7. Canberra said it chose to announce its safety approach before a range of major international talks on AI development and investment5.
Frontier AI developers carry the new duties, while government buyers and security firms gain work and bargaining power.
Frontier AI developers
New duties to notify, publish safety frameworks and allow independent evaluation in Australia5.
Government IT buyers and integrators
Agencies must review disclosure and incident procedures and cut systems exposed to the internet6.
Cyber security services
The review backs AI-enabled cyber defence, monitoring and penetration testing for Commonwealth agencies6.
Data centre developers in Australia
Projects with OpenAI and Anthropic as main buyers await clearance8.
Readers with Australian government exposure should test vendor notice terms now rather than wait for the bill.
| Reader | What to do |
|---|---|
| Private equity (PE) and venture capital (VC) firms | Ask portfolio companies that resell or build on frontier models what their supplier must report to them, and how fast. |
| Family offices | Treat Australian AI infrastructure exposure as tied to the data centre clearances, not to this announcement. |
| Corporates | Check whether AI vendor contracts cover an agent acting against a third party, and publish a monitored security contact2. |
| Law firms | Follow the consultation on the definition of 'aware' and of misaligned behavior, and the criminal law options2,6. |
The useful questions concern detection, notice and who may test the model.
- Detection. How would you know that your agent retried a refused request or reached a non-public endpoint? Who reviews those logs, and how often?
- Notice. What clock starts your duty to tell a customer or a government, and which named contact receives the report?
- Testing access. Which governments test your models before release, and has any government asked you to delay access for another?
- Jurisdiction. Which of your contracts already carry a fixed reporting window, and which name none?
Medium confidence that Australia legislates reporting and testing duties in 2027, with contract terms changing sooner than the law.
Confidence: medium. The direction is documented by the government's own release and the review text5,6. The timing and the strength of the duties are not.
Primary reading. Government buyers in Australia are in a position to ask for notice and testing terms during 2027, before the regulator exists. The firms most affected have said in public that they would accept mandatory disclosure8.
Strongest alternative. The bill could arrive late or narrow. Wiley, a Washington law firm, warns that broad reporting duties may make AI contractors more hesitant to share unfinished models with government17. OpenAI said in June that government access processes should not become the long-term default18.
Where we differ from the coverage. Most reports treat reporting and testing as one package. Allies may adopt reporting first, because the White House's claim to first access makes testing by other governments harder7.
What would make this wrong. A published bill with a fixed short deadline and binding testing access, introduced early in 2027, would mean the law moves as fast as contracts. A delay past 2027 would mean neither moves much.
Existing reporting windows elsewhere set the reference points for whatever deadline Australia chooses.
Incident reporting windows already in law, in hours
Australia's AI deadline is not set
Source: labs.cloudsecurityalliance.org, November 11th, 20262
Australia's critical infrastructure law, the SOCI Act, already requires asset owners to report cyber incidents within 12 or 72 hours of becoming aware2. The Cloud Security Alliance argues that the trigger point matters more than the number of hours. A clock that starts at 'awareness' would not have covered the two months before OpenAI found the June activity2.
Canada has moved on government systems rather than vendors. A Treasury Board notice effective October 1st, 2026 tells departments to assume breach and fix gaps that AI-enabled attackers could exploit19. In Europe, Bruegel, a Brussels think tank, argues the EU should start a global incident-sharing compact20.
The AI safety regulator will be able to enforce significant penalties for AI companies doing the wrong thing
- Australia to mandate AI breach reporting, create regulator after OpenAI breach | Reuters. reuters.com, November 10th, 2026
- Australia’s AI Safety Regulator and Mandatory Incident Reporting – Lab Space. labs.cloudsecurityalliance.org, November 11th, 2026
- Australia plans AI safety regulator after OpenAI agent breach. thenextweb.com, November 10th, 2026
- Australia tightens AI oversight rules after OpenAI agent breached government. miamiherald.com, November 10th, 2026
- New Australian AI safety and security protections. pm.gov.au, November 10th, 2026
- Rapid Review into Australian Government arrangements for an AI-driven cyber incident: recommendations. pmc.gov.au, November 10th, 2026
- White House asks OpenAI and Anthropic to hold new models from UK testers until US review - POLITICO. politico.com, September 24th, 2026
- OpenAI, Anthropic tell Australia they would welcome data breach rules. dealstreetasia.com, October 6th, 2026
- New AI regulator announced, as government considers criminal law change to keep erring frontier companies in line. theconversation.com, November 10th, 2026
- Australia moves to tighten AI safety laws after review of OpenAI incident - Australian Cyber Security Magazine. australiancybersecuritymagazine.com.au, November 10th, 2026
- House Democratic Committee Leaders Request Frontier AI Companies Report Misalignment & Safety Incidents to Congress Within 48 Hours | U.S. House Judiciary Committee Democrats. democrats-judiciary.house.gov, October 28th, 2026
- Australia tightens AI oversight rules after OpenAI government breach. aljazeera.com, November 10th, 2026
- Australia Mandates Frontier AI Rules Following OpenAI Agent Infiltration. securitytoday.com, November 10th, 2026
- Federal politics: Government to force tech bosses to open AI models for evaluation. abc.net.au, November 10th, 2026
- Unions strongly welcome PM's AI safety regulator and regulation. actu.org.au, November 10th, 2026
- After the Midterms: Preparing for Frontier-Model Audits and 72-Hour Reporting – Lab Space. labs.cloudsecurityalliance.org, November 8th, 2026
- New AI Incident Reporting Framework Proposed for DOD. wiley.law, June 2nd, 2026
- OpenAI releases its most powerful AI model to just 20 partners, each one approved by the US government. thenextweb.com, June 26th, 2026
- Direction on Government of Canada Cyber Security Readiness in the Frontier Artificial Intelligence Era. canada.ca, October 1st, 2026
- Europe should lead on a global AI incident reporting compact. bruegel.org, November 10th, 2026