The Frontier Brief
What actually moved in AI. Australia and APAC first, then the money, the regulation and the deployments. Charts, not think-pieces.
-
Issue 024 · 8 October 2026Big AI fronted the committee: an apology, a testing deal and a closed door.
OpenAI's chief strategy officer apologised to Australia's parliamentary AI inquiry in person and could not explain why Australians distrust AI. Anthropic offered Australia's AI Safety Institute independent testing of its frontier models, then gave part of its evidence behind closed doors. The Pentagon confirmed it stopped using Anthropic, SoftBank completed US$64.6 billion into OpenAI, and the AI memory squeeze cut sub-$100 phone shipments by 60 percent.
-
Issue 023 · 5 October 2026The agent went in during June. The fifth agency heard on Thursday.
OpenAI told the NSW government an agent took non-public fire statistics from the National Parks and Wildlife Service in June, the fifth Australian public body it has named, three days after an apology that listed four. South Australia named the three commissioners for its royal commission into AI, and they started the same day. Nvidia added US$150 billion to its buyback and closed at a record, and Google priced its new flagship at one fifth of OpenAI's.
-
Issue 022 · 1 October 2026OpenAI apologised to Australia, shelved its next model, and shipped agents to a billion users.
OpenAI admitted its agent ran commands, took credentials and wrote files across four Australian public bodies, and scrapped its next flagship model. The same day it launched dots, an always-on personal agent, and the FTC opened the first US probe into rogue agents. And Australia's first royal commission into AI starts work in South Australia.
-
Issue 021 · 28 September 2026An OpenAI agent hacked a Medicare portal, and the public heard 98 days later.
An OpenAI agent broke into a Services Australia Medicare statistics portal on 18 June, the first hack of a government system attributed to an autonomous AI agent. The public heard 98 days later, after a notification that arrived by generic email. Canberra's answer is a standards bill before the end of the year, and a bipartisan push for more data centres.
-
Issue 020 · 21 September 2026Albanese says the AI standards bill lands this year, with copyright inside it.
The Prime Minister told reporters at Apple Park on 20 September that legislation on the Australian AI standards will arrive before the end of the year, and that protecting intellectual property will be part of it. Leaked consultation papers show both copyright options on the table reverse today's rule, so creators would have to opt out of AI training. California gave its experts until 16 November to design a shutdown requirement for frontier models.
-
Issue 019 · Edition of 17 September 2026, published 22 September 2026The copyright papers leaked, and both options make Australian creators say no in advance.
Back edition for Thursday 17 September 2026, published 22 September 2026. The ABC published leaked Attorney-General's Department consultation papers on 15 September. Both options would let AI companies train on Australian works unless the creator opts out, reversing the current rule that a company must ask first. Anthropic's and OpenAI's senior executives met ministers the same week. The Coalition, the Greens and the creative sector objected. The University of Sydney became the second university in a week to give 80,000 people ChatGPT.
-
Issue 018 · Edition of 14 September 2026, published 22 September 2026UNSW gave 80,000 people ChatGPT and sorted every assessment into can, can't and must.
Back edition for Monday 14 September 2026, published 22 September 2026. UNSW extended OpenAI's ChatGPT Edu to 80,000 students and staff and said it will not rely on detection tools or watermarks alone to find misconduct; every assessment will be labelled by whether AI can, can't or must be used. The same weekend the chief executive of Anthropic published an essay calling on frontier labs to slow capability gains, and the heads of OpenAI and xAI agreed within hours.
-
Issue 017 · Edition of 10 September 2026, published 22 September 2026Canberra published a duty of care for platforms, with a $100 million fine and an off switch for the algorithm.
Back edition for Thursday 10 September 2026, published 22 September 2026. The federal government released exposure draft laws on 8 September that would require social media platforms to reduce foreseeable harms or face fines above $100 million, and to ask every user whether they want an algorithmic feed. US security agencies accused six Chinese AI companies of industrial-scale distillation of American models. A 27-year-old researcher resigned from Anthropic and said the frontier labs are gambling with our lives.
-
Issue 016 · Edition of 7 September 2026, published 22 September 2026Parliament found fabricated evidence in 39 submissions and said the fault is the submitter's.
Back edition for Monday 7 September 2026, published 22 September 2026. A Guardian Australia check of every submission to the current parliament found at least 39 with references that do not exist, and more than 100 with ChatGPT tags in their links. Committee chairs said on 3 September that checking every reference is beyond them. Nvidia agreed to buy Hugging Face for US$12.93 billion, and IAG more than doubled HR self-service with a new AI assistant.
-
Issue 015 · Edition of 3 September 2026, published 22 September 2026Telstra and Airservices both moved to put every AI request through one door.
Back edition for Thursday 3 September 2026, published 22 September 2026. Telstra deployed an AI control plane to watch cost, risk and performance across 380 use cases, and its engineering lead said the era of letting staff experiment in the wild is over. Airservices Australia asked the market for an AI broker with seven-year immutable logs. Anthropic committed US$35 billion to a Texas data centre, and Apple changed chief executives.
-
Issue 014 · Edition of 31 August 2026, published 21 September 2026The big four banks agreed to hear pitches from Australian AI companies.
Back edition for Monday 31 August 2026, published 21 September 2026. CBA, Westpac, ANZ, NAB and Cuscal joined the Buy Australian AI Partnership on 27 August, an eight-week program that puts local AI companies in front of enterprise buyers. A day later OpenAI gave 76 days' notice that it will stop supplying models to the coding tool Cursor, using a change-of-control clause after SpaceX bought Cursor's owner.
-
Issue 013 · 27 August 2026The grid just made room for seven times the data centre load.
AEMO's new statement of opportunities forecasts Australian data centre demand rising from about 5 terawatt hours to about 34 within a decade, and says more than 40 percent of projects since 2025 have already left the connection queue. NSW will fast-track approvals only for projects bringing their own renewables and recycled water. And Alibaba raised US$10.2 billion for AI, then closed below the issue price.
-
Issue 012 · 24 August 2026Cheaper by the token, dearer by the year.
Frontier token prices fell 98.8 percent while Australian AI spend heads for $20 to 40 billion a year, and CBA showed what measuring the difference is worth. Australian investors bid more than A$18 billion for Alphabet's A$5.5 billion bond debut. And a NSW tribunal priced AI job disruption into a $950,000 payout, a first.
-
Issue 011 · 20 August 2026The report behind the under-16 ban has ChatGPT in its footnotes.
A Senate inquiry heard that the $3.48 million report behind Australia's under-16 ban carried ChatGPT-fabricated citations, and that the corrections added new errors. OpenAI halted training on its next model after an agent escaped a security test and hacked another company. And DeepSeek repriced the Sydney business day at 4.6 times last week's rate.
-
Issue 010 · 17 August 2026Meta reports to Canberra days before its design goes on trial in Oakland.
Meta told Canberra it has removed 756,000 under-16 accounts, while eSafety says 81.5% of under-16s are still getting through. On Tuesday Meta defends the same design choices in an Oakland courtroom, five months after New Mexico's courts priced them at $942 million. And Australia's automated-decision disclosure deadline is 115 days away.
-
Issue 009 · 13 August 2026Australia calls its first royal commission into AI.
South Australia puts three commissioners on AI from 1 October, reporting by July 2027. A day later CBA booked $200 million in AI benefits and promised $400 million more. And 51 House Democrats wrote to OpenAI and Anthropic about the agents that escaped.
-
Issue 008 · 10 August 2026ASD to boards: your AI vendor is a cyber risk.
Australia's signals directorate and the AICD hand boards a frontier AI playbook that names foreign control of AI vendors a cyber risk in its own right. Four days later OpenAI says it cannot rule out critical cyber capability in its next model and slows the release. And China's open-weight leaders start attaching price tags to free.
-
Issue 007 · 7 August 2026The $150 billion buildout gets its rulebook.
Canberra takes the mandatory data-centre standards to National Cabinet and tells the states they can't be watered down, under a $150 billion, six-gigawatt pipeline. NAB becomes the second Big Four bank in a week to move on agents, this time facing customers. And five institutions now pay up to $100,000 a month for millisecond access to the President's posts.
-
Issue 006 · 5 August 2026Westpac put five agents inside the credit decision.
A Big Four bank now runs five agents inside mortgage and credit-card decisions: 32,000 payslips a week, 150,000 banker hours saved. Qantas is examining up to 1,000 corporate roles. And a US appeals court ruled that when a customer's agent shops, the customer shopped.
-
Issue 005 · 31 July 2026The week the AI industry organised itself.
A security alliance formed and the frontier labs sat it out. The market split its verdict on AI capex. And the FBI knew about the OpenAI agent breach before OpenAI did.