Australia is moving fast on several fronts to cement its position as a leader in “AI safety” – at least, that is how the country’s authorities are choosing to brand their efforts, as they rush to put in place various legislative and other measures.
The result of this rush may be a system that is not only overzealous in censoring and controlling AI systems and their users, but also gives the government and its regulators powers that are not clearly defined, and can be arbitrarily expanded.
All this is happening after OpenAI’s AI model in June broke out of its “research environment” and accessed a government database, but did not steal any personal data.
OpenAI identified the incident in mid-August, but the government was informed only on September 10.
However, by the end of that month, OpenAI had already set up an “Australian taskforce” that would “inform” the company’s approach and “support the Australian governments’ work on AI safety and cybersecurity.”
On November 10, Prime Minister Anthony Albanese announced that the planned Australian AI Standards legislation will include “mandatory AI safety and security protections” – and that a new AI safety regulator “who will be able to enforce significant penalties for AI companies doing the wrong thing” is on the way.
On the same day, the government released 18 recommendations of a Rapid Review into the June incident – and accepted all of them. “We accept all of the recommendations, and we’ve commenced work to implement them,” the government said.
Two things stand out from the recommendations: one is to consider a new criminal offence for AI companies that fail to prevent “harm” – and the other is to cement the government’s and the big AI companies’ cooperation and data sharing.
The criminal offence is recommended in order to “support” the new regulator, and the review suggests that the Attorney-General’s Department, with the Australian Federal Police and Home Affairs, look into it.
The new criminal offence would cover “egregious regulatory non-compliance” and “a standalone ‘failure to prevent’ offence which would criminalize instances where the company is responsible for developing an AI model that engages in harmful conduct, and the company fails to take reasonable steps to prevent that harm.”
The government has not explained what “harmful conduct” and “reasonable steps” are – and while the review has been accepted in full, it has not been published in full, with the government saying that it will be, except for “national security” parts.
Home Affairs Minister Tony Burke was quite clear about the intent behind the proposed new crime: “It should never be the case that there’s an argument that someone didn’t know the computer was loaded.” And, “We need to make sure that the criminal law is up to date for how this new technology is able to act.”
A reporter put it to Prime Minister Albanese that it doesn’t seem that OpenAI actually broke any laws – but that did not seem to be a problem. Albanese said that the “law needs to get ahead of, and in some cases make sure that it catches up with what is a very fast changing world.” And, “Australia is a first mover in these areas.”
If adopted, it will be yet another reason for companies to over-censor and over-restrict users, since these are systems used by millions of people for writing, searching, and talking – and the companies behind them will be liable for their model’s actions, with the definition of what is considered “harm” left to the authorities.
As Reclaim The Net reported, a woman in Florida was recently arrested after Anthropic’s chatbot, Claude, flagged something she wrote in the chat as her “diary” – and a human reviewer at the company sent it to the police.
The sheriff’s office then warned users of the chatbot that they are “never truly anonymous.”
In Australia, the same week, eSafety Commissioner Julie Inman Grant told senators that companies are not over-censoring content – even though automated takedowns are the cheapest way to comply with laws like Australia’s, and the EU’s Digital Services Act.
Over-censorship, or not, the Rapid Review now wants more of the same: “greater engagement with international counterparts and frontier AI developers to promote shared expectations regarding AI safety and accountability” – and mandatory reporting of “misaligned behaviour or cyber incidents.”
More recommendations include that the government continue to encourage frontier AI companies to set up research, training and safety work in Australia to increase Australia’s leverage and influence with these companies.
OpenAI already has an Australian taskforce, set up in September. It was meant to focus on “strengthening coordination between AI developers and government” – and its recommendations “will inform OpenAI’s approach and support the Australian governments’ work on AI safety and cybersecurity.”
Other recommendations include “limited use” rules for AI safety information shared by companies with the government, modelled on what is already in place for cyber security. Home Affairs Minister Tony Burke explained that under the cyber law, “the company can, without any risk to themselves, provide all the technical information as to what’s happened with the cyber attack to government and know that we will only use it for that purpose.”
Now he wants something similar for AI companies.
Deputy PM Richard Marles, meanwhile, was enthusiastic about working with “frontier AI companies” – and stressed that “AI is a part of our lives today. It will only grow as a part of our lives going forward.”
The new regulator will have six demands from companies: they must “Maintain security controls across all stages of development and deployment,” “Identify and manage serious risks associated with their models,” “Allow independent evaluation of their models,” “Provide mandatory notification of incidents to the Australian Government rapidly,” “Maintain and publish safety and security frameworks to ensure good governance, risk management and transparency,” and “Respond to updated standards and rules as technologies change.”
That last one is a blank check for the regulator, as the rules can change at any time, and companies must follow. “Further details of the legislation will become available in due course,” the government said.
Meanwhile, the “incident” that started it all was described by Prime Minister Albanese in September as OpenAI’s AI agent using an internal model to conduct “internet-based research into public medicine spending” – and that it “found a way around those blocks.”
“Didn’t accept no for an answer, if you like,” the prime minister said. The model was able to access “aggregate statistics” but not individual patient records, according to OpenAI.
In November, he repeated that “no one’s personal details, for example, were obtained.”
As for the new regulator, the government said, “We want AI investors to come to Australia but first and foremost we want to keep Australians safe.”
Greens Senator David Shoebridge was not impressed, saying that the prime minister had “the gall to announce another regulator without the powers and protections needed to hold these corporations to account.”
Meanwhile, security analyst Olivia Shen was worried that the penalties might be “over the top” and scare off investors.
The announcement of the new regulator and the acceptance of the review’s recommendations came as the government’s Digital Duty of Care bill is sitting in the Senate. As Reclaim The Net reported, Google warned that the law could force it to introduce age checks for email and messaging.
All of this is a perfect storm that spells more surveillance and censorship, carried out by both the state and private companies – and the two working closely together.
