What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
iTechGuides is reader-supported. When you buy through links on our site, we may earn an affiliate commission. As an Amazon Associate I earn from qualifying purchases. Learn more
Not conclusively. Khan Academy has built a layered safety system around Khanmigo, its GPT-4-based tutor: content moderation, usage limits, red-team testing, feedback channels, educational-use rules, and adult visibility of children’s activity. Those measures make Khanmigo more constrained than an unconfigured general chatbot, but public evidence does not prove that every harmful response is caught, every answer is correct, or that the controls reliably produce long-term learning.
What Khanmigo’s guardrails are designed to do
Khan Academy announced Khanmigo as a GPT-4-powered AI tutor and teaching assistant in 2023. The company says it adds tailored prompts, product rules, monitoring, and other mitigations rather than exposing students to a raw general-purpose chatbot.
Moderation and escalation
Khan Academy’s safety help page, updated July 1, 2026, says moderation technology looks for interactions that may be inappropriate, harmful, or unsafe. When moderation is triggered, the company says an email is sent to an adult connected to the child’s account. It also describes user feedback and appeal channels, and says an account may be disabled for violations.
The organization warns that AI can produce harmful or inappropriate material and that Khanmigo can be wrong. Its responsible-AI disclosure uses the plain wording: “AI can be incorrect or misleading.”
#1 Best Overall
Limits intended to reduce misuse
Khan Academy says daily usage limits are part of the design because longer conversations can increase the chance of poor behavior. Fine-tuning and prompt engineering steer the system toward tutoring and teaching tasks. Terms and in-product messaging prohibit non-educational use and attempts to jailbreak the system.
Red teaming and feedback
The company says it conducts red-team exercises to search for vulnerabilities and reviews user feedback to identify problems. It follows risk-evaluation practices adapted from the National Institute of Standards and Technology and the Institute for Ethical AI in Education. Khan Academy also acknowledges that current safeguards cannot eliminate every risk.
Who can use Khanmigo, and who can see a child’s activity?
Access rules are part of the safety model, although Khan Academy says they can change. Individual registrants must be at least 18. Minors may use Khanmigo through a parent- or guardian-linked child account, a district partnership, or an assigned Writing Coach essay activity.
For child accounts, Khan Academy says the child is told that chat history and activities can be visible to parents or guardians and, where applicable, teachers and school administrators. Adults can review chat logs through the adult dashboard. The company also says shared images are not stored under its privacy policies.
Rank #2
These are published product and policy descriptions, not an independent measurement of how consistently alerts, visibility, retention rules, or account actions operate in practice. They should not be treated as a complete privacy audit.
How Khan Academy evaluates risk
Khan Academy’s published framework rates risks by likelihood and impact, then lists mitigations for high-priority cases. For inappropriate or harmful use, its examples include OpenAI’s Moderation API, responses that point users to community standards, adult notifications, possible account disabling, transcripts visible to parents and teachers, red teaming, and rules against non-educational use and jailbreak attempts.
At the March 2023 launch, Khan Academy estimated that these mitigations would reduce the example risk rating from high to medium. The company explicitly noted that the initial ratings were estimates made before a conversational product had been tried. It later reported that many inappropriate interactions involved children testing limits and that conversations often stopped after a flag. That is a company account, not an independently published incident analysis.
Recommended Free Tools
“Enough” depends on the risk being measured
A single safety label hides several different questions. The public record supports the following distinctions:
Rank #3
| Question | What is documented | What is not established publicly |
|---|---|---|
| Can Khanmigo reduce harmful or non-educational interactions? | Moderation, usage limits, red teaming, account controls, adult alerts, and educational-use rules are described by Khan Academy. | A verified rate for missed harmful content, jailbreak success, false negatives, or total safety incidents. |
| Can students trust its answers? | Users are warned about factual and mathematical errors. A specialized math agent introduced in 2026 verifies calculations and expressions, while math-error rates are monitored. | A comprehensive independent benchmark of Khanmigo’s factual or mathematical accuracy. |
| Does tutoring improve independent work? | Product tests track cognitive engagement, premature answer-giving, and correctness on the next same-skill question without Khanmigo assistance. | Proof that short-term gains generalize to durable learning or that safeguards caused those gains. |
| Is child activity accountable and private? | Child-account visibility, adult dashboards, moderation alerts, and an image-retention statement are published. | A comprehensive independent audit of child-safety controls, data handling, or implementation rates. |
What Khan Academy’s recent product tests show
In a May 2026 product report covering roughly six months of testing from October 2025 through April 2026, Khan Academy said it measured response latency, correctness on the next same-skill problem without Khanmigo help, and cognitive engagement classified as passive, active, or constructive. It also monitored premature answer-giving, math-error rates, and interactions per thread.
The company said it deployed a change when its estimated “chance to win” exceeded .95 and no guardrail metric showed a negative impact. Across more than 15 million tutoring threads, it reported:
- A 3.4% improvement in next-item correctness across 608,000 tutoring threads after a summary of recent learner performance was added.
- A 2.7% improvement across 1.36 million threads when the system surfaced unmastered prerequisite skills.
- A reported 6.1% combined improvement from those two changes.
These figures are Khan Academy’s 2026 company-reported A/B-test results. They indicate iterative measurement of learning-related behavior, not a safety audit, a failure rate for moderation, or evidence that every student will retain the material long term. Khan Academy said a fuller paper on the metrics, infrastructure, and experiments would be presented at the 27th International Conference in AI for Education.
What independent studies say about learning
Two-year middle-school experiment
The working paper “One Click Away: AI Tutoring with Khanmigo in a Two-Year School Experiment,” by Philip Oreopoulos and Nina Low, describes a cluster-randomized study in 18 middle schools in Hamilton County, Tennessee, during the 2024–25 and 2025–26 school years. Students were below grade level and used Khanmigo during existing math intervention periods. The configuration was intended to coach rather than simply provide answers.
Khan Academy’s August 2026 summary says the study was not designed or run by Khan Academy. It reports an approximately 0.06-standard-deviation combined two-year intent-to-treat estimate, 0.08 standard deviation in year two, and 0.14 standard deviation in a secondary analysis of students who remained in the intervention throughout year two. Students used Khanmigo infrequently, and the comparison condition included Khan Academy and other existing tools. These results are evidence about a broader school intervention, not proof that Khanmigo’s guardrails caused the gains or that they are sufficient for child safety.
Undergraduate lunar-phase study
A 2025 peer-reviewed mixed-methods study by Nedim Slijepcevic and Ali Yaylali involved 69 undergraduates learning lunar-phase concepts. It compared Khanmigo with Google search, with a paper-only group emerging during the experiment. Learning improved across conditions, but the authors found no statistically significant outcome difference between groups. Participants valued Khanmigo’s step-by-step guidance and personalization while viewing it as supplementary rather than a replacement for instruction. The small sample, short exposure, and quality of printed materials limit what can be inferred.
Why broader GPT-4 studies do not answer the Khanmigo question
A PNAS study of GPT-4 interfaces in a Turkish high-school mathematics setting found that an unguarded or differently prompted tutor can affect learning in problematic ways. It did not evaluate Khanmigo, Khan Academy’s moderation stack, or its operational safeguards. Findings about a generic GPT-4 tutor cannot be transferred directly to Khanmigo.
How OpenAI’s model statistics should be interpreted
OpenAI’s 2023 safety reporting said GPT-4 was 82% less likely than GPT-3.5 to answer requests for disallowed content and 40% more likely to produce factual content. OpenAI also described Khan Academy as a developer partner using tailored mitigations on top of default model safeguards.
Those are model-level comparisons. They do not measure Khanmigo’s complete product stack, student conversations, moderation misses, privacy controls, or educational effectiveness. It would be incorrect to call Khanmigo “82% safer” or “40% more accurate” on that basis.
What remains unproven
Publicly available materials do not provide a verified Khanmigo safety-incident count, moderation false-negative rate, jailbreak success rate, or comprehensive independent audit of child-safety controls. The absence of a public tally does not establish that no incidents occurred.
The evidence is strongest for describing Khan Academy’s intended controls and its recent product-level measurements. It is weaker for independently measured failure rates, privacy implementation, long-term learning transfer, and the relationship between any specific safeguard and student outcomes.
Outdated Drivers Are Slowing You Down
One free scan finds every outdated or missing driver and matches the right update for your exact hardware.Free scan · exact hardware matchWindows Errors? Fix Them Before They Spread
Repair common Windows errors and clear accumulated junk for a smoother, more stable PC - no reinstall needed.Free scan · no reinstallA practical standard for families and schools
- Use an accountable account. Confirm whether a child is using a parent-, guardian-, school-, or district-linked account and who can review activity.
- Keep an adult in the loop. Treat moderation alerts and dashboards as additional oversight, not as a substitute for supervision.
- Verify important work. Check factual and mathematical answers, especially when the result affects grades, health, safety, or a high-stakes decision.
- Look for reasoning, not just answers. A useful tutor should prompt explanation and practice without assistance, rather than encourage copying.
- Ask what evidence supports a deployment. Distinguish vendor-reported tests from independent studies, and ask whether outcomes were measured on unaided follow-up work.
- Review privacy terms. Check current retention, visibility, and data-sharing language because product policies and access conditions can change.
“We think GPT-4 is opening up new frontiers in education… It’s transformative and we plan to proceed responsibly with testing to explore if it can be used effectively for learning and teaching.”
Kristen DiCerbo, Khan Academy chief learning officer, quoted in OpenAI’s March 2023 announcement
Khanmigo’s safeguards are meaningful risk mitigations with human-oversight features, but the public record does not justify calling them foolproof or conclusively sufficient. Whether they are “enough” depends on the consequence being considered—and on continuing independent measurement of failures as well as successes.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
