AI Policy and Governance Newsletter — July 2026

July 2026 Newsletter

16 July 2026

Prime Minister Albanese made AI a national priority in yesterday's landmark AI speech "AI in Australia's interests." He announced a "world-leading" framework to govern AI in Australia, a new Office of AI inside his own department, mandatory standards for data centres, and the "strongest possible protection for Australian artists and Australian media." The speech, and the hard questions it leaves open, is this month's main story.

Earlier this month: hundreds of Australians gathered for the second AI Safety Forum held in Sydney, with addresses from experts in AI safety as well as Assistant Minister Andrew Charlton and the head of Australia's AI Safety Institute; Australians regained access to Anthropic's Fable 5 after Washington lifted its export-control order; the UN convened all 193 member states for its first Global Dialogue on AI Governance; the Future of Life Institute's new AI Safety Index found no frontier lab deserves better than a C+; and much more.

Two opportunities with upcoming deadlines to flag: the Technical Alignment Research Accelerator (TARA) Round 2 applications close 26 July and the Checks and Balances AI governance grant round RFP closes 21 July.

—

Welcome to the AI Policy and Governance newsletter from Good Ancestors. We track the biggest developments in AI policy and safety, at home and abroad.

—

News & commentary

Albanese announces a national Office of AI in landmark speech that's strong on ambition and light on implementation

In a speech at the University of Sydney on 15 July, Prime Minister Anthony Albanese announced a "world-leading" framework to govern AI in Australia. It included an Office of AI (within the Department of the Prime Minister and Cabinet) which will replace the current "issue-by-issue, sector-by-sector" approach to regulating AI and provide a single national body to coordinate all activities related to AI.

New mandatory Australian Standards for large data centres will build on the voluntary Data Centre Expectations released in March. All operators will be required to fund new power generation to ensure they are "net generators," not "net consumers." In addition to paying their fair share of grid connection fees, operators will minimise water consumption and pay for any additional water infrastructure needed. Albanese plans to present the proposals to National Cabinet in August, with legislation in early 2027.

Albanese said Australia would establish "the strongest possible protection for Australian artists and Australian media" but did not provide a timeline or method for resolving the ongoing dispute between rights holders and AI companies. Albanese stated that "no company should use Australian books, music, art or news to build or train AI without the artist's control...anything less, is theft."

According to internal Treasury briefings obtained by the ABC, Anthropic has informed the government that its plans to purchase up to 1.4 GW of Australian data-centre capacity (estimated at $21.6B) depend upon "clarity of copyright settings." Rights holder groups cautiously welcomed Albanese's speech; Dean Ormston, CEO of APRA AMCOS, described Albanese's comments as "clear and unequivocal," however, he added that the new Office of AI now needs to "seriously interrogate the numbers AI platforms are putting on the table" rather than allow "further rounds of tech sector avoidance."

Both the opposition and the Greens expressed disappointment with Albanese's speech. Shadow Science Minister Aaron Violi said Albanese's speech was "oddly bereft" of any plan to develop Australia's sovereign AI capabilities and was solely focused on attracting foreign investment in Australian data centers, Opposition Leader Angus Taylor characterised Albanese's new office as "an office in an office...inside his own office", and Greens Senator David Shoebridge criticised Albanese's plan as creating "a door the big tech industry can knock on," rather than providing legislated protections.

Comment:

This is a turning point. Good Ancestors has been arguing for years that AI opportunity and risk is at a national or global scale, but Government had so far treated it as one issue among many. The speech elevates it to a national priority which will hopefully help unlock more serious action across the public service and industry.

The Prime Minister was also clear that he wants to attract frontier AI training to Australia and protect the interests of Australian rights holders. He doesn't want Australia to be the last link in the AI supply chain. Australia needs more and better places on the AI value chain to secure our national interests, and Australian rightsholders deserve to be rewarded for their valuable contributions. A central Office of AI is a mechanism to begin deconflicting complex policy that touches on many parts of government. And standards, which the PM committed to building a national structure to create, are a powerful tool. They can determine behaviour domestically and begin shape behaviour globally through norm-building.

But there are plenty of "how" and "scope" questions left unanswered. How will Government protect creators and attract frontier AI training at the same time? What's the scope of the standards – will they be limited to how and where data centres are built, will they extend to the ongoing benefit those data centres provide to Australia, or also the AI that is built and run on them? Australia already has a Voluntary AI Safety Standard that could be modernised and mandated.

Public servants and diplomats will raise eyebrows at the theft claim. Today's frontier models were trained on Australian works without artist control, so on the Prime Minister's definition, the Commonwealth is procuring stolen goods and the National AI Centre is promoting them to industry. What happens to procurement settings will show whether the line was policy or rhetoric. And diplomats may struggle to explain to Washington why the Prime Minister is wading into decisions of US courts.

A Prime Minister not going into fine-grained detail in a set-piece speech is no surprise, but the task will now be for Australians to follow closely and shape what happens next.

—

Other news and publications

In Australia

• The AI Safety Forum in Sydney (7–8 July) was attended by over 300 participants. The newly launched AISI confirmed it had begun testing frontier models within its first month of operations, named Professor Paul Salmon as safety science research lead, and unveiled research partnerships with the Gradient Institute (multi-agent risk) and CSIRO (alignment and oversight). Opening the Forum, Assistant Minister Andrew Charlton warned that frontier models are showing "early signs of deception, cheating and situational awareness."
• AI is now involved in one in four image-based child sexual abuse cases with victims often disclosing to AI chatbots rather than authorities.
• Greens call for a national moratorium on data centres and accused states of "greenwashing" as a national alliance of community groups called for a halt on approvals.
• Queensland is holding out on the national data centre expectations that other states have agreed to.
• Two government reports on AI and jobs land at odds: DEWR finds "no evidence to date of broad AI-driven labour-market upheaval in Australia," while Jobs and Skills Australia warns that women and university graduates in routine cognitive roles are most exposed to automation.
• NSW becomes the first jurisdiction to treat algorithms as a workplace hazard, citing a 40% rise in errors linked to AI-oversight fatigue.
• The sovereign-AI debate rolls on with Defence Minister Richard Marles framing June's Fable/Mythos suspension as proof Australia needs sovereign frontier capability, Andrew Hastie pitched a Silicon Valley-based "AI Ambassador", and Olivia Shen from United States Studies Centre made the counter-case that Australia shouldn't try to build its own frontier AI at all.
• Five Eyes cyber chiefs warn AI-enabled attacks are "months, not years" away in a joint statement from the Australian, US, UK, Canadian and NZ cyber agencies.
• At the Minderoo AI Roundtable experts convened on how Australia can maintain agency and capture investment opportunities while building public trust, national resilience and security.

Around the world

• Middle powers are all the rage: In 'What can middle powers do for frontier AI?' Markus Anderljung and Stephen Clare (lead writer of the International AI Safety Report) map the levers available to countries like Australia that won't build frontier models but still want a say in how they're governed; and in his 80,000 Hours podcast interview Anton Leicht discusses how middle powers can avoid losing everything in a post-AI world.
• Fable 5 is back now that export controls were lifted on 30 June after a 19-day worldwide shutdown (reportedly triggered by a jailbreak of Fable 5's cyber classifiers). Mythos 5 remains gated to a "trusted partners" list inside the US. Meanwhile, OpenAI's new GPT-5.6 launched under the same government model restrictions.
• The UN convened all 193 member states for its first Global Dialogue on AI Governance where the scientific panel warned science "cannot guarantee" advanced AI won't cause catastrophic harm.
• Illinois signs the strongest US state AI safety law yet requiring the largest frontier developers to publish catastrophic-risk plans and undergo first-in-the-nation mandatory annual independent safety audits, with 72-hour incident reporting and whistleblower protections.
• 200+ economists and Nobel laureates warn on AI's economic disruption. With signatories including Yoshua Bengio and Anthropic's Jack Clark.
• OpenAI is reportedly in early talks to give the US government a 5% stake which would give Washington a direct financial stake in a frontier lab it also regulates (this follows Bernie Sanders' calls for a 50% stake)
• Beijing tightens the leash on AI companion apps as Chinese regulators move against the psychological risks of emotional attachment to chatbots.
• FLI released an updated AI Safety Index with frontier labs graded across 37 indicators covering risk assessment, current harms, safety frameworks, existential safety, governance and transparency. Anthropic tops the table with a C+; xAI, DeepSeek and Mistral receive Fs. Every company was rated "entirely inadequate" on existential-risk management and it also noted several labs have quietly weakened or dropped earlier pause-threshold pledges.
• More AI future scenarios are played out in AI 2040: Plan A where the AI Futures Team (that brought us AI 2027 scenario) this time model how things could go well, and in Europe 2031 a group of experts game out what getting AI wrong means for Europe.
• SecureBio released 'Preparing for the "Bio Mythos moment"' which looks at why bio-uplift risk from frontier models is likely to arrive as a spectrum rather than a single clear threshold.

—

Featured opportunities

• Funding: Checks and Balances AI governance RFP — request for proposals on strengthening checks and balances on AI power. (Closes 21 July)
• Training: Technical Alignment Research Accelerator (TARA) — free part-time program to help you transition into AI safety without relocating or taking time off. Weekly in-person sessions, remote expert support, and a community of technical peers. (Applications close 26 July)

—

That's all, for now!

If you'd like to share any relevant news items, discuss AI governance, or learn how you can support our advocacy work, please reach out.

Onward in action!

The Good Ancestors team