AI Limitations & Disclaimer
What this platform does
Bizarus helps you organise, structure and interrogate research: forming questions, recording searches, capturing evidence, laying out arguments, surfacing contradictions and drafting summaries of material you have supplied. Some of that assistance is produced by AI language models.
Its purpose is to help you think more clearly and work more systematically. It is a workspace and a set of instruments. It is not an authority.
What this platform does not do
It does not verify facts. It does not confirm that a source exists, that a quotation is accurate, or that a study says what it is described as saying. It does not check that a reference is real.
It does not provide professional advice of any kind, and does not evaluate the quality of your research, the validity of your method, or the defensibility of your conclusions. It does not replace your supervisor, your ethics committee, your institution's guidance, your discipline's standards, or your own judgement.
It does not know things. It produces text that is statistically likely given its training and your input, which is a different activity from knowing, and the difference matters precisely when the output is most persuasive.
How AI output fails, specifically
These are not hypothetical risks. They are documented, recurring properties of the technology, listed so that you know what to look for rather than being told vaguely to be careful.
Fabricated sources. A model can produce a reference with a plausible author, title, journal, year, volume, page range and DOI, for a work that has never existed. The format will be correct. The source will not be there.
Misrepresented real sources. More dangerous than invention, because it survives a existence check. A real paper is cited for a claim it does not make, or its finding is reversed, overstated, or stripped of the qualification that made it defensible.
Confidence unrelated to correctness. Output does not become more hesitant as it becomes less reliable. The tone of a fabricated citation is identical to the tone of a correct one. You cannot use fluency, certainty or detail as a signal of accuracy.
Staleness. Models are trained to a cutoff date. Retractions, corrections, superseded guidance, changed law and later findings after that point may be invisible to them. A confidently stated position may have been abandoned by the field.
Silent omission. A summary can be accurate in everything it says and misleading in what it leaves out: the contradicting study, the limitation section, the sample size, the conflict of interest.
Numerical and statistical error. Models manipulate numbers as text. Effect sizes, confidence intervals, percentages, sample counts and conversions can be wrong in ways that look entirely ordinary.
Inherited bias. Training data reflects what has been published, in which languages, by whom, and what has been indexed. Output can quietly reproduce those imbalances and present a partial literature as the literature.
Inconsistency. The same question asked twice can produce different answers, and a model may agree with a correction that is itself wrong. Agreement is not verification.
Retrieval failure. Where the system searches external databases, a search can return nothing relevant, miss the key work, or surface a weak source. An absent result is not evidence that nothing exists.
Human review is required, not recommended
This platform is built on the principle that AI assists human reasoning and does not replace it. That principle is also a condition of use.
You must independently verify every factual claim, citation, quotation, statistic, date and attribution before relying on it. Open the source. Read enough of it to confirm it says what it is cited as saying. Do not cite anything you have not read. Do not submit anything you have not checked.
Where the stakes are higher, the obligation is higher: work that will be published, submitted for assessment, relied on by others, used in a decision affecting a person, or presented as evidence requires proportionately more scrutiny, not less. Time pressure is not an exception to this and will not be treated as one.
Nothing on this platform is a second opinion, a peer review, an ethics check, or a substitute for a qualified human being in the loop.
Nothing is published or accepted automatically
The system never promotes a finding into your knowledge base, publishes content, or accepts a conclusion on your behalf. Where automated gathering surfaces candidate material, it is staged for review and a person decides. That design is deliberate, and it means the responsibility for what you accept is, and remains, yours.
Domains where you must not rely on this platform
Output must never be used as the basis for a medical, clinical, diagnostic or treatment decision; for legal advice or a legal position; for financial, investment or tax decisions; for safety-critical engineering or operational decisions; or for any decision materially affecting another person's health, liberty, finances, employment or welfare.
If your research touches these areas, the platform may help you organise literature about them. It must not be the thing you rely on. Consult a qualified professional in the relevant field.
Third-party AI providers
AI output is generated by third-party providers, currently Google Gemini, Groq and OpenRouter, as described in our Privacy Policy. We do not control their models, training data, behaviour or changes to them. A provider may alter or withdraw a model at any time, which can change output quality without notice to us or to you. We make no warranty in respect of any third-party model.
No warranty
All content and output is provided "as is" and "as available". To the fullest extent permitted by law, Bizarus AI (Mauritius) excludes all warranties, conditions and representations, whether express, implied or statutory, including any implied warranty of accuracy, completeness, currency, reliability, merchantability, fitness for a particular purpose, or non-infringement.
No advice or information, whether oral or written, obtained from the platform creates any warranty not expressly stated here. Using the platform creates no advisory, professional, supervisory or fiduciary relationship between you and us.
Limitation of liability
To the fullest extent permitted by law, we are not liable for any loss or damage arising from your use of, or reliance on, the platform or its output. That includes, without limitation: academic penalties, allegations or findings of misconduct, failed examinations or vivas, rejected submissions, retractions, corrections, loss of funding, professional or disciplinary consequences, reputational harm, loss of data, loss of profit, and any indirect or consequential loss, whether or not we were advised such loss was possible.
This applies however the loss arises, including where output was inaccurate, incomplete, fabricated, outdated, biased or misleading, and including where the platform was unavailable or lost data.
Stated honestly, because a term that overreaches is more likely to be struck out than one that does not: nothing here excludes liability that cannot lawfully be excluded, including liability for death or personal injury caused by negligence, for fraud or fraudulent misrepresentation, or any other liability that applicable law does not permit to be limited. Where a limitation is not permitted where you live, our liability is limited to the minimum the law allows.
This notice applies whether or not you have read it
This disclaimer governs your use of the platform in full, from your first use, whether or not you have read it, and whether or not you were shown it. It is published, permanently accessible, and linked from every page of this site. Not having read it is not a defence and does not reduce its effect.
By using the platform you acknowledge that you have been given notice of the limitations described here; that you understand AI output can be confidently and undetectably wrong; that you accept the obligation to verify independently before relying on anything; and that you accept responsibility for what you choose to do with the output.
If you do not accept these conditions, do not use the platform.
Errors, corrections and changes
Content may contain errors, become outdated, or omit relevant information despite genuine effort. We may correct, update or remove content at any time without notice, and are under no obligation to do so. We may change this notice as the platform and the underlying technology develop; the date at the top shows when it last changed.
Severability
If any provision of this notice is held unenforceable in any jurisdiction, it is severed to the minimum extent necessary and the remaining provisions continue in full force. A provision held unenforceable in one jurisdiction remains effective in every other.
Contact
Questions about this notice can be sent through the contact page. If you find an inaccuracy in published content, please report it: that is how it gets fixed.
About this notice
This notice forms part of our Terms of Service and should be read with them and with the Privacy Policy.
It is written to be understood rather than to be impenetrable. That is deliberate: a warning nobody can follow protects nobody, and a limitation a reader could not reasonably have understood is worth less, not more.
Issued by Bizarus AI (Mauritius).