Australia should be given priority access to the most capable artificial intelligence tools developed in the US and the Trump administration was wrong to temporarily withhold access to the powerful Claude Mythos tool.
asserted
administration → give → tool
That’s the view of Lori Trahan, the Democratic congresswoman for Massachusetts, helping to lead a push for more stringent regulation of AI in the US.
asserted
That → ’ → US
She believes that as part of the push for better “rules of the road” to ensure the technology is properly controlled, the US should also foster more open collaboration with its Five Eyes intelligence partners, particularly Australia and the UK.
asserted
US → believe → partners
“Innovation and safety are not at odds with each other, and I think we should just remind ourselves that when we set those standards high, we bring the rest of the world with us.
asserted
we → think → us
But that’s also incumbent on our collaboration with our allies,” she said.
asserted
she → ’ → allies
In April, Anthropic released its Claude Mythos AI model, which is particularly skillful at finding and exploiting vulnerabilities in IT systems.
asserted
which → release → systems
Australian government security agencies and even Australian citizens working on AI in the US – along with all other foreign nationals – were initially barred from working with Mythos after a Trump administration edict, which was partly wound back after a couple of months.
asserted
which → work → months
Anthropic, which makes the Claude AI chatbot and is on the verge of a record $USD1.4 trillion initial public offer, has heavily restricted access to Mythos to financial institutions, major tech companies and cloud services providers like Amazon.
asserted
which → make → Amazon
Trahan, who is a member of Congress’ House Committee on Energy and Commerce, told this masthead the technology should not have been withheld from Five Eyes intelligence partners.
asserted
technology → tell → partners
“We rely on those relationships with our allies, and so I think that there’s a collaboration that we need to strike.
asserted
we → rely → that
In fact, we should have a Western stack [a collection of software tools] that is unified and fostering collaboration,” she said.
asserted
she → have → collaboration
Speaking at the Takeoff AI conference in Washington DC, Trahan said “there’s no question that AI was on the ballot” during the recent mid-term elections and that exit polls showed “that AI mattered in the congressional races”.
asserted
AI → speak → races
There is a growing push in the United States for greater AI regulation and safeguards and Trahan, working alongside California Republican Jay Obernolte to achieve bipartisan reforms, has helped lead the way.
asserted
Trahan → be → way
The Australian government recently announced an AI Safety Institute in response to a push for greater regulation of artificial intelligence domestically, something which has been anathema to the Trump administration.
asserted
which → announce → administration
Trahan thinks the position of the Trump administration could shift now the Democrats have a majority in Congress.
uncertain
Democrats → think → Congress
“You never know what deal the president will sign,” she said.
asserted
she → know → deal
David Wroe, director of AI security the organisation Good Ancestors, welcomed Trahan’s comments and said Mythos had the power to “carry out any cyber activity you can imagine”.
“Obviously, we want Australia to be first in line to have access to the most advanced US technology, but I understand why US companies and government want to be careful about who has access, given how powerful they are becoming,” he said.
asserted
he → welcome → access
“Lori is right to say we need to spread risks, the US needs to look at building infrastructure in other trusted countries, and it is very much in Australia’s interest to have that done in Australia.
asserted
that → say → Australia
Australia is one of a handful of countries being examined by Anthropic as a second location to set up AI training data centres, which are more powerful than centres that simply answer users’ questions to chatbots.
asserted
that → examine → chatbots
Training centres can be located more remotely from major population centres but the question of what the most advanced models train, and compensation for copyright holders in Australia and overseas, remains unresolved.
asserted
models → locate → Australia
James Massola travelled to Washington DC as a guest of the Australian Council on AI Strategy.
asserted
Massola → travel → Strategy