AI in Schools
Brussels writes the rubric ACER will turn into PISA 2029
Digital Attitudes, 29 June 2026
The Monday read is one international rubric that has just walked behind the desk of the Camberwell research organisation that will turn it into the assessment 90 countries sit, one cybersecurity model whose US government release list has been confirmed at more than 100 critical infrastructure organisations while Australian institutions stay off the page, and one AI teammate that has just shifted from a private back and forth into a shared channel level coworker.
EU and OECD ship the 19 competence AI rubric ACER will turn into the PISA 2029 test
The European Commission and the OECD published the joint AI Literacy Framework on Wednesday 24 June at the European Digital Education Hub flagship event in Brussels, with 19 competences spread across four domains (engaging with AI, creating with AI, managing AI, and shaping AI). Development drew on more than 100 countries through surveys, written reviews and focus groups, more than 2,000 participants in total, teachers representing 41 percent of survey respondents, and the framework was built with support from CodeAI and an international expert group. The next operational step the document names: this framework will inform the first OECD international measurement of AI literacy, the PISA 2029 Media and Artificial Intelligence Literacy (MAIL) assessment, the cycle that the Australian Council for Educational Research (ACER) was appointed to lead from Camberwell on 1 April. Pia Ahrenkilde Hansen, the European Commission's Director-General, on the day: "Preparing young people for a world shaped by artificial intelligence starts in the classroom." EdTech Innovation Hub on the EU and OECD setting 19 AI literacy competences for schools, 24 June.
Why this matters: For state Department of Education curriculum leads and AISWA, ISV and Catholic education AI policy authors writing the term 3 AI competency expectations document, the rubric the Australian school system's term 3 work will eventually be assessed against has just been published, and the question the next paper answers is whether the school's own AI literacy framework already maps to the four domains the AILit framework names or whether the next rewrite cycle adopts the EU and OECD wording wholesale before the 2027 NAPLAN cycle starts the conversation under the same labels. For NESA, the Victorian Curriculum and Assessment Authority and the South Australian Certificate of Education authorities, the operational read is that the PISA 2029 MAIL instrument ACER will deliver across 90 countries now has its design blueprint, and the question the year 9 and 10 curriculum review answers is whether the digital technologies strand picks up the engage, create, manage and shape four domain split before the 2029 sample year ten cohort sits the test. For ACER's own product portfolio (the PAT suite, the Reading Comprehension Test, the schools improvement work), the framework is the document the Camberwell team will pivot the next two and a half years of MAIL item writing against, and the question the next ACER funding submission answers is whether the institutional capability that delivers PISA at the international scale also gets paid to localise the same instrument back into the Australian national assessment suite.
Mythos 5 clears the US government bar for 100 critical infrastructure organisations
US Commerce Secretary Howard Lutnick signed off on Saturday 27 June on a limited redeployment of Anthropic's Mythos 5, the cybersecurity focused AI model first headlined in April as the company's most capable and the one it had initially declined to release publicly, after a fortnight of daily talks between Anthropic and federal officials. The cleared list runs to more than 100 US organisations: federal agencies, the operators that defend critical infrastructure, and a handful of major corporations whose risk classification met the new release framework. Mythos 5 is the model whose April demonstration runs reportedly turned up thousands of high severity vulnerabilities across every major operating system and web browser autonomously, and the 12 June Trump administration export freeze (the one that took Anthropic's Fable 5 access off the Australian table for ten days, the Monday brief a week ago) was framed at the time around the cybersecurity uplift the model gives anyone holding it. NBC News on the US government green light for the limited Mythos 5 re-release, 27 June.
Why this matters: For the Australian Cyber Security Centre and the Australian Signals Directorate, both of whom carry the country's critical infrastructure protection mandate against the same Five Eyes AI cyber threat clock the Tuesday brief priced a week ago, the procurement question is whether the Mythos 5 access ticket is now on the AUSMIN agenda for the next Australia US strategic dialogue, or whether the model stays an inside US perimeter capability while ASD's developer skills register works out how to defend against the same vulnerabilities the same model finds. For UNSW College and the other Australian universities standing up cyber defence applied bachelor's enrolment pipelines, the curriculum read is that the leading defensive AI tooling the next cohort would normally train on is now behind a US government clearance gate, and the question the term 3 curriculum committee answers is whether the syllabus pivots to the cleared OpenAI GPT-5.5-Cyber and Codex Security combination Anthropic does not own. For Australian university vice chancellors and DVC research leads watching the Anthropic Claude for Education contract surface (Sydney, ANU, RMIT), the operational read is that the model behind the Education tier and the model behind Mythos 5 are now formally on different distribution paths, and the institutional research procurement question is whether the next contract conversation with Anthropic mentions Mythos at all.
Anthropic ships Claude Tag into Slack as the persistent channel teammate
Julie Bort at TechCrunch reported on Tuesday 23 June that Anthropic has shipped Claude Tag, an in Slack agent that any member of a channel can summon by typing @Claude, into general beta for Claude Enterprise and Claude Team customers. The new product replaces the existing Claude in Slack application on a 30 day migration window and ships with four behaviours the previous app did not carry: multiplayer interaction where one Claude instance serves the whole channel, persistent memory built from following the channel's conversations, ambient initiative where Claude surfaces relevant updates without being asked, and asynchronous task execution where Claude can pursue projects over hours or days. Anthropic disclosed that 65 percent of its own product team's code is now created by an internal version of Claude Tag, and the enterprise controls administrators get on rollout cover token spend limits at organisational and channel level, plus a full audit log of which user triggered which agent action. TechCrunch on Anthropic's Claude Tag learning the company one Slack message at a time, 23 June.
Why this matters: For Australian university IT directors and CIOs at the Sydney, UNSW, Melbourne, Monash and Queensland clusters where Slack is the staff and HDR student communication backbone, the operational read is that the AI assistant inside the Slack tenant is no longer a private DM between a single staff member and the bot, and the question the next academic governance paper answers is whether the staff channel that has Claude Tag enabled becomes a system of record (audit log, persistent memory, the agent's working notes left visible to the channel) before the institution's own records management policy catches up. For school business managers and IT leads at the independent and Catholic systems that have run Slack as the office and operations layer for a couple of years (the Edmund Rice, MLC and Wesley networks among them), the procurement read is that the AI tooling the operations team has been using as an individual aid has just become a channel level coworker the principal will see in the audit log, and the question the term 3 budget paper answers is whether the school renews the existing Claude in Slack tier or moves to the Enterprise tier the new tool requires. For Australian ed-tech vendors selling AI assistants into school and university Slack tenants (Slate, FYI, the assorted homegrown bots), the read is that Anthropic has just shipped the canonical pattern of how an AI agent shows up in a shared channel, and the next vendor pitch deck either accepts the pattern or explains why a different distribution shape works better than the one that has 65 percent of Anthropic's own product team's code behind it.
The Monday thread is that the assessment rubric, the cybersecurity tooling and the staffroom AI have all moved upstream this weekend, the AI literacy framework has walked into ACER's PISA build pipeline, the most capable cybersecurity model has been gated to a US critical infrastructure list, and the Slack channel level AI has shifted from a single user DM to a multiplayer coworker. More on the operational read at digitalattitudes.com.au.