Study Hall Blog

Category: AI, Universities and Careers

  • Safe, Responsible AI at College CoPilot

     

    Protecting student voice, professional judgment and private information

    Responsible AI helps people make better decisions without taking over their thinking, crossing academic-integrity boundaries or exposing information to the wrong people. 

    Dakota gives tutors and counselors faster and deeper preparation. It enables you to give students personalized support designed to comply with the Common App Fraud Policy while helping ensure each student’s work stays authentically their own.

    Five Commitments Across the Platform

    Data privacy College CoPilot limits data use to educational purposes, protects student records and does not expose private information across unrelated users or contexts.

    Research-based Pedagogy Dakota does not think for the student. It asks questions, explains, prompts reflection and supports basic revision such as grammar while the student supplies the ideas, judgment and substantive work. Dakota follows defined learning frameworks and supports understanding rather than answer-getting. 

    Strong Guardrails Dakota does not create prohibited work, fabricate student information or help users evade academic and institutional rules.

    Controlled Access Roles determine access and action. Dakota understands whether it is supporting a student, parent, tutor or counselor and limits information and assistance accordingly.

    Workspace Control Schools, counselors and tutors turn any workspace on or off at any time, controlling which support is available.

    How Safety Works in Each Workspace

    Test Prep

    • For tutors: Dakota summarizes performance over time, prepares session priorities and creates targeted practice. It saves analysis time while the tutor retains responsibility for diagnosis, instruction and progression.
    • For students: Dakota teaches through mistakes, prompts students to identify the traps they fell for and adapts practice. It explains reasoning rather than simply supplying answers and measures progress across repeated evidence—not one correct response.

    College Research & Thrive

    • For counselors: Dakota connects the Thrive Report and college research into a clear evidence base for discussion. It supports list-building without diagnosing the student, declaring a college objectively “best” or predicting admission.
    • For students: Dakota frames focused research questions and assignments. It cites sources, distinguishes verified facts from inference and leaves the student to investigate, reflect and decide.

    Application Coach

    • For counselors: Dakota surfaces the most important questions, structural issues and revision priorities. It streamlines early feedback while the counselor retains the judgment and relationship that shape authentic work.
    • For students: Dakota helps students understand prompts, deepen reflection, organize their own material and correct objective mechanics. It does not supply substantive ideas, invent experiences, rewrite passages, replace the student’s vocabulary or voice, or generate submit-ready work.

    Every student affirms: “I understand that the ideas, experiences, meaning and substantive wording in my applications must be my own.”

    Privacy, Security & Compliance

    • College CoPilot uses role-based permissions and purpose-limited data practices so each user receives only the information and tools appropriate to their role.
    • School deployments are designed to support FERPA and COPPA obligations and applicable GDPR and UK GDPR requirements. Documented controls cover access, correction, retention, deletion, subprocessors and incident response.
    • Security controls cover authentication, encryption, access, logging, backups and data handling by AI providers. College CoPilot keeps these claims tied to current product controls and contractual commitments.
    • Dakota is designed around the Common App authorship boundary. Its guardrails help prevent substantive AI-written application content, fabricated experiences or credentials, and attempts to disguise AI authorship.

    Responsible AI Leadership

    Sofia Fenichell, founder of College CoPilot, serves with the EdSAFE AI Industry Alliance and on its Transparency Committee and contributes to wider responsible-AI initiatives. This work informs College CoPilot’s commitment to safe, accountable, fair and effective AI in education.

    Policy References

    Common App Fraud Policy

    U.S. Department of Education: FERPA

    Federal Trade Commission: COPPA Guidance for Ed Tech

    European Commission: Principles of the GDPRpan>

  • Michigan’s LSA Honors Prompt Isn’t Testing Writing. It’s Testing How You Think.

     

     

    Most colleges are trying to keep AI out of admissions essays. The University of Michigan’s Honors Program is requiring it.

    Their 2026 LSA Honors prompt:

    Use an AI text generator to write an essay on the topic “Why is the LSA Honors Program a good fit for me?” Submit the AI-generated essay unchanged. Then write your own essay analyzing it — where did the AI get it right, where did it get it wrong, and why?

    For the third year, Michigan’s LSA Honors Program is handing students AI and saying: now show us you can think for yourself.

    The Shift No One Is Talking About

    The conversation around AI in college admissions has centered on detection — can schools tell if a student used ChatGPT? Michigan reframed the question.

    They’re not asking: Did you use AI?

    They’re asking: Can you tell when AI is wrong — and why?

    That’s what the workplace will demand from this generation.

    The student has to read the AI essay closely, spot the generic claims, find where the logic is thin or the writing follows a formula — and articulate all of it in 750 words.

    Why This Matters Beyond Michigan

    A 2025 Kaplan survey found that 50% of admissions officers view AI use in applications unfavorably. Most schools are still focused on prohibition and detection. Brown bans AI in applications entirely. BYU threatens to rescind admission. Yale calls it “application fraud.”

    Michigan chose a different path: test the skill, not police the tool.

    AI-generated essays aren’t bad because they’re “cheating.” They’re bad because they’re generic. A 2025 Cornell study found that AI essays sound competent but feel hollow — they lack the specifics and self-awareness that make a personal statement personal.

    The Connection to Everything Else

    It’s about what kind of thinker your teen is becoming.

    AI is already handling first drafts across industries. The analyst who accepts an AI-generated market summary without questioning its assumptions is less valuable than the one who catches the flawed reasoning. The lawyer who submits an AI-drafted contract without spotting the ambiguous clause is a liability.

    Read carefully, evaluate whether it holds up, explain why it doesn’t — that’s what separates people who use AI as leverage from people who get replaced by it.

    It’s also what the SAT measures. The ability to read dense, unfamiliar material and make precise judgments under constraint.

    Michigan just made that connection explicit. The question is whether your teen is building that capacity now.

     

  • Spotlight: Brown University in the Age of AI

     

     

    While headlines about artificial intelligence in higher education tend to focus on bans, plagiarism scandals, and hand-wringing, Brown University has spent the past nine months doing something far more interesting: building an institution-wide AI infrastructure from the ground up — and appointing the leadership to steer it.

    For families navigating the college admissions landscape, Brown’s approach signals an important shift in where elite education is heading. The university isn’t treating AI as a problem to manage. It’s treating it as a core institutional capability.

     

    A New Role With Unusual Scope

    In July 2025, computer science professor Michael Littman stepped into a newly created position: Associate Provost for Artificial Intelligence. The role covers AI education, research, and responsible practices, with a charge that extends into operations and external engagement — a remit that Littman himself believes is unusually broad.

    “Other institutions have hired people to look after one or two of these areas, but I don’t know of any others that are unifying all of them in a single role,” he told Brown’s newsroom shortly after starting. He reports directly to Provost Francis J. Doyle III and leads a cross-university coordinating committee that connects academic initiatives with administrative functions.

    Littman is no bureaucratic hire. He previously served as Division Director for Information and Intelligent Systems at the National Science Foundation, where he oversaw $200 million in annual research funding and co-authored the 2023 update to the U.S. AI Research and Development Strategic Plan. He won the AAAI Outstanding Educator Award in 2024. And he earned his PhD at Brown in 1996, making his return to College Hill both a homecoming and a statement of intent.

     

    Is B-Minus the New F? What AI Means for Brown’s Academic Culture

    One of the most provocative ideas to emerge from Littman’s early work is a direct challenge to how universities grade student work. Referencing the book Teaching with AI, which his committee read together, Littman laid out a striking argument: if AI can produce C-plus or B-minus work on virtually any assignment, then perhaps that level of output simply isn’t good enough anymore.

    “Maybe a B-minus is the new F,” Littman said. “So now we have all this headroom above that where the student and the AI can work together to produce something better.”

    For Brown families, this raises a question worth thinking about — even if the university itself hasn’t posed it yet. Brown is famous for its flexible grading culture: the Satisfactory/No Credit option, no failing grades on transcripts, and a philosophy that encourages intellectual risk-taking over grade anxiety. If AI raises the baseline of what constitutes acceptable work, what does that eventually mean for a system built around generous definitions of “satisfactory”?

    To be clear, Littman hasn’t called for changes to Brown’s grading system, and the university hasn’t publicly connected his comments to S/NC policy. But the underlying tension is real: when the floor of competence rises, the institutions that have historically given students room to experiment may need to rethink what that room looks like.

     

    Free AI Tools for Every Student

    In late 2025, Brown rolled out university-wide access to AI tools — a move that puts practical technology directly into students’ hands rather than leaving them to cobble together their own subscriptions.

    Students now have free access to Google Gemini Chat and NotebookLM through their Brown accounts, cleared for the university’s highest data safety classification. The Office of Information Technology also launched Transcribe, an AI-powered transcription tool built internally through Brown’s AI Sandbox project, which Littman noted performs better and costs significantly less than commercial alternatives.

    “We think these tools are extremely relevant to the business of universities,” Littman told the Brown Daily Herald. “They shouldn’t just be available to people on campus who have their own personal resources to pay for them.”

    The OIT is also developing Librechat, a service that will consolidate multiple large language models into a single chatbot interface, giving students and faculty access to several AI models through a single Brown-managed platform.

     

    $20 Million for AI and Mental Health

    In July 2025, the National Science Foundation awarded Brown a $20 million grant to lead a new national AI institute focused on developing trustworthy AI assistants for mental and behavioural health applications. The AI Research Institute on Interaction for AI Assistants (ARIA) brings together researchers from Brown, Dartmouth, Carnegie Mellon, NYU, UC Berkeley, and several other institutions.

    Led by Brown computer scientist Ellie Pavlick, ARIA aims to build AI systems capable of context-aware, sensitive interactions with people — particularly those in states of distress. The work addresses what Pavlick calls the hardest open problems in AI today: understanding cause and effect, modelling internal human states, and building systems transparent enough to earn trust.

    “Any AI system that interacts with people, especially those who may be in states of distress or other vulnerable situations, needs a strong understanding of the human it’s interacting with,” Pavlick said.

     

    Why Interdisciplinary Matters More Than Ever

    Littman makes a compelling case that AI literacy is no longer a tech skill — it’s a research skill that cuts across every discipline. In his Q&A with Brown’s newsroom, he gave a vivid example: particle physicists are now generating more data per second than they can possibly store, so they’re using AI to filter in real time for what’s scientifically interesting. The same pattern is emerging in biology, public health, economics, environmental science, and beyond.

    This is where Brown’s open curriculum becomes a genuine strategic advantage. A student could combine cognitive science and computer science, which is exactly the intersection where the ARIA mental health institute operates. They could study how AI reshapes public policy through Brown’s new data science master’s programme, which focuses specifically on AI governance. They could explore what AI means for the humanities through the Cogut Institute’s “Models-Scale-Context” lab. Brown isn’t siloing AI into the computer science department. It’s threading it through neuroscience, engineering, applied math, public health, philosophy, and the arts.

    “AI enables people to do science more effectively,” Littman said. “We need people to be educated about what the possibilities are in AI, how it can be used, what kind of questions it can answer, and what it can’t answer.”

     

    Leapfrogging Into the Workforce

    Perhaps the most important signal for families came in Littman’s comments about the job market. He was candid: AI chatbots are already well-suited to do the kind of work companies expect from new hires, and some employers are slowing entry-level hiring as a result.

    His answer was constructive.  “Institutions like Brown need to find a way to help their graduates leapfrog over these entry-level positions and start contributing at a higher level,” he said.

    He described current AI as having “broad competence” but still lacking “deep insight.” That distinction matters. The students who thrive in the next decade won’t be the ones who can use AI tools because everyone will be able to. They’ll be the ones who understand what questions to ask, what the limitations are, and how to apply AI responsibly within a specific domain. That’s exactly what an interdisciplinary education at a place like Brown is designed to produce.

     

    What It Means for Families

    Brown’s AI strategy reflects a broader truth about where top universities are heading: the schools that will best prepare students for the next decade aren’t the ones banning AI — they’re the ones teaching students to use it critically, creatively, and responsibly.

    For prospective applicants, Brown’s investment in AI infrastructure, research leadership, and classroom integration is worth noting. It suggests an institution that sees AI fluency not as a threat to liberal education, but as an essential part of it.

     

    Sources: