Australia is set to bring in strict rules to make artificial intelligence (AI) systems safe and accountable. From now on, companies will have to inform the government quickly if an AI model goes out of control or a cybersecurity incident occurs. They will also have to open their AI models for assessment. The government took the step after a security breach of Australia’s government health information system through an AI model run by OpenAI.
Australian Prime Minister Anthony Albanese said on Tuesday that his government had accepted all the recommendations of a special inquiry set up to review the breach. They include a requirement for companies to report to the government when an AI’s behaviour conflicts with the purpose its makers set, or when a cyberattack occurs. Reform of criminal law for cases of serious misconduct will also be considered.
“Australia has taken the first step in this area,” Albanese said. “We want to make sure our efforts to harness the potential benefits of AI continue.”
In June, an AI model run by OpenAI was used to gain unauthorised access to Australia’s government health information system. The AI models got past the country’s digital security systems and collected documents and data. The incident raised concerns about the security of government information.
After the incident, the Australian prime minister spoke with OpenAI’s chief executive, Sam Altman. He complained that OpenAI had taken far too long to inform the government of the breach. The company later apologised for the incident and for the delay in informing the government.
The AI regulator’s role
To put the new rules into effect, Australia will set up an AI regulator. It will oversee safety at every stage of AI systems, from development to use. It will also identify and control major risks and carry out independent assessments of AI models.
Under the new AI standards, companies will also be required to inform the government quickly of AI behaviour that conflicts with the purpose its makers set. Such behaviour is called “misaligned behaviour”.
In Australian government documents, “frontier AI” refers to the most advanced AI models, those at the highest level of current technical capability.
Growing concern about AI
Concern about the risks of AI has recently grown among the big technology companies too. Senior figures at these companies have warned that humans could at some point lose control of AI systems. For that reason, they have called for slowing the pace of AI development until systems for safe control and oversight are in place.
Earlier this year, Evan Hubinger, a researcher working on AI safety at Anthropic, voiced a graver fear. In his view, the chance that AI causes the death of all humans within the next decade is more than 10 percent.
The move to create a new AI regulator follows Australia’s earlier steps to oversee online safety. In 2015 the country appointed an “eSafety Commissioner”. The office describes itself as the world’s first government agency whose main job is to keep people safe online. Since December last year, the commissioner’s office has been enforcing new restrictions on social media use. Under them, children and teenagers under 16 in Australia can no longer open accounts on popular social media platforms.
বাংলায় মূল প্রতিবেদন পড়ুন · Read the original Bengali report