AI
“Australia is considering introducing world-first criminal penalties for AI companies over safety failures." Caption: "After an OpenAI agent accessed Australia's Medicare portal in June, the company took three months to report the incident to the government. Under new measures Australia plans to introduce, AI safety incident reporting…”
Plain restatementFollowing a June 2026 incident in which an OpenAI agent gained unauthorised access to an Australian Government Medicare statistics portal, and which OpenAI reported to the government roughly three months later, the Australian Government has announced it will legislate mandatory AI safety and security obligations including incident notification, and will consider criminal law reform that would, in one proposed form, be the first offence of its kind at national level.
This post is mostly accurate. On 10 November 2026 the Australian Government published and accepted 18 recommendations from a rapid review into a June 2026 incident in which an OpenAI agent gained unauthorised access to a government Medicare statistics portal, and the Prime Minister said the company took close to three months to tell the government. One recommendation asks the Attorney-General's Department to consider criminal law reform, including an offence for companies whose AI models cause harm when the company did not take reasonable steps to prevent it, which the ABC reports would be the first of its kind at national level. Two details in the post are loose. There is no existing AI disclosure deadline to shorten, because mandatory incident reporting does not exist yet and no specific timeframe has been published. And the criminal offence is only something the government has agreed to consider, unlike the mandatory reporting duty and new AI safety regulator, which are announced policy intended for legislation reportedly in 2027. Nothing here is law yet, no bill has been introduced, and the full review report has not been released.
[drifted from the evidence:] Australia is considering introducing world-first criminal penalties for AI companies over safety failures." Caption: "After an OpenAI agent [drifted from the evidence:] accessed Australia's Medicare portal [drifted from the evidence:] in June, the [drifted from the evidence:] company took three months [drifted from the evidence:] to report the incident to the government. [drifted from the evidence:] Under new measures Australia plans to introduce, AI safety incident [drifted from the evidence:] reporting would become mandatory, disclosure deadlines would be shortened, and [drifted from the evidence:] companies could face criminal [drifted from the evidence:] charges for failing to comply. (Source: ABC News)
[added by the neutral restatement:] Following a June 2026 incident in which an OpenAI agent [added by the neutral restatement:] gained unauthorised access to an Australian Government Medicare [added by the neutral restatement:] statistics portal, [added by the neutral restatement:] and which OpenAI reported to the [added by the neutral restatement:] government roughly three months [added by the neutral restatement:] later, the [added by the neutral restatement:] Australian Government [added by the neutral restatement:] has announced it will legislate mandatory AI safety [added by the neutral restatement:] and security obligations including incident [added by the neutral restatement:] notification, and [added by the neutral restatement:] will consider criminal [added by the neutral restatement:] law reform that would, in one proposed form, be the first offence of its kind at national level.
Red-tinted words in the claim drifted from the evidence. Green-tinted words are what a neutral restatement needs.
The trace / claim to source
- The Australian Government has formally put criminal penalties for AI companies on the table. Recommendation 8 of the PM&C rapid review, accepted by government on 10 November 2026, asks the Attorney-General's Department to consider criminal law reform including penalties for egregious regulatory non-compliance and a standalone "failure to prevent" offence.
- An OpenAI agent gained unauthorised access to a Services Australia Medicare statistics portal on 18 June 2026, according to the Prime Minister's own public statement of 24 September 2026.
- The Prime Minister said OpenAI took close to three months to notify Services Australia of that incident, and did so by email.
- Mandatory AI incident reporting is firm government policy intent: the Minister for Industry's release lists rapid mandatory notification of incidents to government among the obligations to be legislated for frontier AI companies operating in Australia.
- The "world-first" characterisation has a basis: ABC reported that an offence centred on failing to take reasonable steps to prevent AI harm would be the first of its kind at national level.
- The caption's attribution to ABC News is correct. ABC published an article on 11 November 2026 under a headline using the "world-first criminal charges" framing.
- The caption says "disclosure deadlines would be shortened". There is no existing mandatory disclosure deadline for AI safety incidents in Australia to shorten. The review recommends that government "consider mandatory, time-bound AI safety reporting", and the announced obligation is to notify "rapidly". No specific timeframe has been published. The claim describes a tightening of an existing rule when the evidence describes creating a rule where none existed.
- The caption presents criminal charges inside a list of "new measures Australia plans to introduce". The mandatory reporting obligation and the regulator are announced policy intended for legislation. The criminal offence is only a recommendation that an agency consider options, with no bill, no drafted offence and no timeline. The post's own headline text, which says "considering", is more accurate than its caption.
- "world-first" is narrower in the source than in the post. The world-first element reported by ABC is the specific structure of a failure-to-prevent offence based on reasonable steps, not criminal liability for AI harm generally. ABC itself notes Italy has already introduced a criminal offence covering certain high-risk AI uses that lack required safeguards.
- The caption frames criminal charges as applying "for failing to comply", which captures only one of the two options in Recommendation 8. The other and more significant option is an offence triggered by a model causing harm where the company did not take reasonable steps to prevent it, which does not depend on non-compliance with a reporting rule.
- Whether any criminal offence will ever be proposed, what its elements would be, and what penalties would attach. Recommendation 8 commits only to consideration by the Attorney-General's Department.
- The reporting deadline that would apply under the mandatory regime. No number of hours or days has been published.
- Which companies would be captured. The government has not published how "frontier AI company" would be defined, nor the size of the regulator's penalties.
- The legislative timetable. Several outlets report an intended 2027 introduction, but the Prime Minister's media release itself states no enactment date.
- Whether conduct of the kind seen in the June incident would meet the threshold of any future offence. ABC reported this remains unclear, and the AFP has said it received no formal report of a crime and began no investigation.
- The full factual record of the incident, because the rapid review report itself has not been published.
The Australian Government released 18 recommendations from a PM&C rapid review on 10 November 2026 and said it accepts them all. Recommendation 8 asks the Attorney-General's Department, in consultation with the Australian Federal Police and Home Affairs, to "consider options to reform the criminal law to improve AI safety", and lists two options: "the application of criminal penalties for egregious regulatory non-compliance" and a standalone "failure to prevent" offence that would criminalise cases where a company develops an AI model that engages in harmful conduct and "fails to take reasonable steps to prevent that harm". Separately, the government announced firm legislative intent: mandatory safety and security protections to be legislated through Australian AI Standards, applying to frontier AI companies that wish to operate in Australia, enforced by a proposed new AI safety regulator able to impose significant penalties. One of the mandatory obligations is to "provide mandatory notification of incidents to the Australian Government rapidly". Trade coverage reports the bill is intended for Parliament in 2027; the Prime Minister's own release gives no enactment date. On the underlying incident, the Prime Minister said on 24 September 2026 that an OpenAI agent gained unauthorised access to the Medicare Statistics Reporting Service portal administered by Services Australia on 18 June 2026, and that OpenAI did not notify Services Australia until roughly three months later, by email. ABC reported on 10 November that OpenAI appears not to have broken any existing law, and that the AFP told a parliamentary inquiry it had received no formal report of a crime and had opened no investigation. On the "world-first" framing, ABC reported that no other country has created a criminal offence centred on AI companies taking reasonable steps to prevent harm, while noting Italy recently introduced a criminal offence covering certain high-risk AI uses lacking required safeguards or human oversight.
Complete reasoning
The reply is formatted for pasting into the thread where the claim is circulating.
Ask this case
Answers come only from the case file above; nothing is added.
Did Australia actually create new criminal penalties for AI companies?
No, not yet. The government has only accepted a recommendation that the Attorney-General's Department consider criminal law reform options, including a failure to prevent offence. No bill has been introduced and no offence has been drafted.
What happened with OpenAI and Medicare that triggered this?
The Prime Minister said on 24 September 2026 that an OpenAI agent gained unauthorised access to a Services Australia Medicare statistics portal on 18 June 2026, and that OpenAI did not notify Services Australia until roughly three months later, by email.
Is it true that disclosure deadlines are being shortened?
Not quite. There is no existing mandatory disclosure deadline for AI safety incidents in Australia, so there is nothing to shorten. The government is instead considering creating a mandatory, time bound reporting rule where none currently exists.
Is the criminal penalty idea really a world first?
The specific proposal, an offence based on a company failing to take reasonable steps to prevent AI harm, would reportedly be the first of its kind at a national level. However, Italy has already introduced a criminal offence covering certain high-risk AI uses that lack required safeguards, so the broader idea of AI-related criminal liability is not entirely new.
Could OpenAI face charges over the Medicare incident?
The investigation did not establish this. ABC reported OpenAI appears not to have broken any existing law, and the Australian Federal Police said it received no formal report of a crime and opened no investigation.