Australia’s AI Safety Institute Has $29.9 Million Over Four Years
An Australian safety group co-founder says one computing contract he signed is worth more than 1.5 times the budget of Australia's AI Safety Institute. The comparison came during a San Francisco study tour where experts gave sharply conflicting forecasts on advanced AI.
Australia's Artificial Intelligence Safety Institute has a budget of $29.9 million over four years, according to an Australian engineer who co-founded a US-based AI safety group. One computing contract his own group signed is bigger than that.
Karl Berzins, co-founder of the San Francisco-based not-for-profit FAR.AI, said the group recently signed a computing deal worth around $50 million over three years. That is more than 1.5 times the institute's entire budget, he said.
A tour with conflicting forecasts
The comparison emerged during a study tour of the San Francisco Bay Area. The Australian Council on AI Strategy, a newish not-for-profit funded by philanthropists, paid for the trip. Participants included heads of Australian foreign affairs and economic think tanks, government officials, scientists and a journalist.
The group visited frontier AI companies, including OpenAI, maker of ChatGPT, and Anthropic, maker of Claude. Journalists were excluded from the Anthropic meeting. Participants also met AI safety experts.
The forecasts they heard split sharply. The optimistic version is a global golden age. The pessimistic version centers on "recursive self-improvement", when frontier models build better versions of themselves without human help or oversight. AI already does most of that work, though not all of it.
What the pessimists say
Pessimists in the discussions put the chance that humanity is wiped out within a couple of years at 10 to 50 per cent. One safety expert summed up the mood: "if you have things you want to do in your life, do them by 2027".
Another participant said "multiple Manhattan Projects" are being run by private companies right now. The US leads the world on AI, several months ahead of China, but regulation there is thin.
Australia's position
Berzins said the federal government understands the risks, and that the belated announcement of the safety institute confirms it. He said the Albanese government is not spending nearly enough.
"My worry is whether we can move fast enough," Berzins said. He added that Australia traditionally likes to play it safe, and "that just won't work here".
Bringing AI training onshore from a company such as Anthropic would give Australia leverage, though it is not guaranteed, according to the account of the tour.
