Unfixed Newsletter — June 11–July 7, 2026
Editors’ note
The tension between the AI labs, government oversight, and national-security review has dominated the last four weeks. OpenAI and Anthropic both ran into federal limits on model access, and the fight is no longer theoretical: the government is beginning to influence which models get released, who can use them, and under what conditions. Colleges are building teaching, research, advising, and administrative workflows on systems whose availability can be changed by companies, regulators, or governments with little warning. That is a campus governance problem, even when the immediate fight is happening far away from campus.
1) Federal government starts shaping access to frontier models
OpenAI restricted access to its GPT-5.6 Sol model to government-approved customers after a Trump administration request. Anthropic’s Fable 5 and Mythos 5 models were caught in a related cybersecurity review. The story moved again this week: the administration lifted restrictions on Anthropic’s Fable 5, while the more powerful Mythos 5 remains limited to select U.S.-based organizations approved by the government.
This follows a June 2 executive order that creates a voluntary framework for federal review of “covered frontier models” before public release. The order says it does not create a mandatory licensing or preclearance regime. But the OpenAI and Anthropic cases show that government review can still shape model access in practice.
Why this matters: Colleges are being encouraged to build around frontier AI systems. Faculty are using them for writing, coding, research support, class preparation, and student-facing work. Institutions are buying enterprise licenses, developing AI literacy programs, and exploring agentic workflows. This episode is a reminder that those systems are not stable public utilities. Access can change because of security politics, export concerns, model-risk reviews, or government pressure.
https://apnews.com/article/anthropic-fable-mythos-trump-claude-028db5135128fce6b38c873bf9cb5e09
2) Canvas moves deeper into AI detection
Instructure has named Copyleaks as an exclusive AI and text-matching partner for Canvas. Under the agreement, Instructure’s global sales organization can offer Copyleaks directly to K-12 and higher-ed institutions. The announcement frames the partnership as a simpler path for schools that want AI detection and plagiarism tools inside the LMS environment they already use.
Why this matters: This is the type of change in the defaults that often ends up shaping decision making. When detection is built into the LMS, academic integrity work becomes easier to routinize and harder to separate from ordinary teaching workflow. That can help faculty who are managing large volumes of student writing, but it also risks pushing campuses toward a tool-first response to an assessment problem.
https://copyleaks.com/learning-management-systems/canvas-plagiarism-checkerhttps://help.copyleaks.com/s/article/Canvas-Asset-Processor-LTI-1-3
3) Students are using AI, but they are not sold on it
Inside Higher Ed’s Student Voice data complicates the simple story that students are either enthusiastic AI adopters or principled resisters. In a May survey of 1,038 two- and four-year college students from 203 institutions, students reported using AI for coursework while also expressing concern about dependence, careers, and inconsistent institutional responses. Two in five students said they were explicitly concerned about dependence on AI tools, while three in five saw AI’s main value in college as learning support.
Why this matters: Students are using AI regardless of policy clarity. The useful faculty response is clearer course-level guidance: when AI is allowed, when it is not, what must be disclosed, what kinds of help count as learning support, and what work students must still be able to do without the machine. Students are asking for norms. Silence is also a policy, and usually a bad one.
4. Cal State faculty push for labor protections around AI
CalMatters reports that the California Faculty Association is backing legislation that would restrict California State University from replacing faculty work with generative AI. The story connects this labor push to CSU’s broader adoption of AI tools, including its large ChatGPT agreement, and to an unfair labor practice charge filed after the system’s AI rollout.
Why this matters: This is a new development in a story we have already been tracking. If an institution buys AI tools for everyone, faculty need to know whether those tools are being framed as optional support, required infrastructure, assessment machinery, advising replacement, course-production automation, or future labor savings. These are not the same thing. Campuses that treat AI as a procurement decision without labor governance will keep running into faculty resistance.
5. The evidence on AI and learning keeps getting messier
A new open-access study in Humanities and Social Sciences Communications surveyed 861 students from STEM and social-science programs in Pakistan, China, and Finland about generative AI use, self-regulated learning, technological self-efficacy, and cognitive offloading. The findings are mixed. Students perceived AI use positively, and the study reports links among AI use, technological confidence, cognitive offloading, and self-regulated learning. The same study also flags student concerns about over-reliance, cognitive overload, metacognitive laziness, and erosion of academic skills.
Why this matters: AI can support learning when students use it as a scaffold, but the same convenience can hide the work students need to practice. That distinction should shape assignment design.
Faculty do not need to ban every use of chatbots to take cognitive offloading seriously. They do need to decide where struggle belongs in the course. If students use AI to generate ideas, check explanations, practice questions, or test their understanding, the tool can support learning. If they use it to skip reading, outsource synthesis, or produce finished work they cannot explain, the assignment no longer measures what it was meant to measure.
Link: https://www.nature.com/articles/s41599-026-07924-3
From Our Work
Zach wrote about the Palo Alto AI-cheating lawsuit and what it shows about institutional confusion around AI detection, academic integrity, and policy consistency. https://www.meltsintoair.org/chatgpt/a-cheating-scandal-and-warning
Nik noted that U.S. News and World Report highlighted Unfixed in its “One New Thing” series, recommending the podcast for listeners following AI and higher education. https://www.meltsintoair.org/chatgpt/us-news-article-unfixed
We also released two new Unfixed episodes: one on why AI is not just another edtech tool, and one with Allison Vaughn and D.J. Hopkins on what AI literacy should mean for faculty, staff, and students. https://www.meltsintoair.org/unfixedpodcast/ep-32-ai-is-not-edtech