Aiberry

The module with the published price is not the one the model runs.

Where the model sits, it sits completely. Smart AI MindCare is a voice and visual assessment conducted by an animated assistant: the interview is administered by a model, the responses are read by an ensemble reading text, audio and video together, and the output is entirely model generated. Strip the machine learning out of that product and nothing remains, not even a questionnaire, because it deliberately refuses the questionnaire format. The model occupies both ends of the interaction rather than scoring an artifact a human collected.

The company sells something else first. Digital MindCare is the tier with a published figure, at $5,000 a year, and what it contains is twelve digital mental health screenings covering depression, anxiety, drug and alcohol use and eating disorders, delivered anonymously with an analytics dashboard, resource routing and expedited crisis referral. No voice assessment and no conversational assistant appears in it. That is a screening delivery and analytics platform built on established instruments, and it is plainly something a buyer would pay for with every model removed, because the company prices it that way.

Measured at the lowest published tier, the artificial intelligence is the engine of a module customers step up to rather than the thing the platform is.

The qualification runs the other way too. Everything distinctive about this company lives in the model: the peer reviewed validation, the multimodal claim, the reason a buyer would choose it over a form. The voice assessment is a core module and not a feature layer.

How oversight works is described rather than asserted, but the threshold inside it is not.

Both named products are anonymous with no sign in, so there is no clinician between the model and the person by default, and the company has built the controls around that rather than around a reviewer. What is published: the product explicitly does not replace counseling or therapy and is positioned as an early supportive touchpoint; users receive personalized insights with relevant resources and suggested next steps drawn from the support options their own organization offers; routing to appropriate support and expedited crisis referral are named product features; and non clinical organizations receive only aggregated, de identified insights, never individual level responses. Stating what the organization does not get is as much a control as stating what the user does.

The Smart AI MindCare tier adds a full transcript of every response, which is a real oversight feature wherever a clinician is involved, because it lets a reader inspect what the person actually said rather than accepting a number.

What is missing is the operating point, and here it is a specific question rather than a generic one. Nothing states what result triggers the expedited crisis referral, what the person sees at that moment, or how a referral is expedited for a user who by construction has no identity, no account and no relationship with the organization. Nothing states whether a threshold exists at all, or what a school, employer or correctional facility is required to have in place before going live.

Anonymity is a privacy virtue, and it is in tension with duty of care. The company has now published a mechanism that acknowledges the tension. It has not published what triggers it.

The method is documented in a journal, and almost none of it reaches the buyer facing material.

What is disclosed holds up. Three modalities are named and the approach is described as an ensemble of machine learners aggregating responses across text, audio and video from a single bot administered conversation. The output is specified precisely: an overall risk score, symptom level insight, and a transcript of every response. Publishing the transcript is itself a transparency feature, because it lets a reader inspect the input rather than only receiving the verdict. The peer reviewed paper documents the methodology, so a technical reader has somewhere real to go.

The gap sits between the paper and the site. The research page claims clinical equivalence to the gold standard and describes the platform as the only clinically validated multimodal mental health assessment platform, without publishing a single figure. No accuracy, correlation, discrimination or error number appears on any buyer facing page. A prospect is asked to accept an equivalence claim on trust when the number exists and is published elsewhere, and to accept an exclusivity claim with nothing supporting it at all.

The company's own citation compounds it. The research page's study link points to a PsyArXiv preprint at psyarxiv.com/btz6n rather than the 2024 journal article, so a reader following the vendor's link reaches an unreviewed version of work that has since cleared review.

No model card, architecture description, operating point or threshold is published, no version or update cadence is stated, and nothing addresses whether third party models underpin the transcription, speech or video components.

Not one component of the chain is identified, and the product makes clear that a chain exists.

No cloud or infrastructure provider, base or foundation model, sub processor register, third party component list or position on whether customer data contributes to model development is published.

The inference available from the product is why this reads as an omission rather than a simple absence. A transcript of every response is an advertised feature of the voice tier, so automatic speech recognition is unavoidably in the chain, and it is a component very rarely built in house. The animated conversational assistant that conducts the interview implies a speech or dialogue layer. Something processes video. Each is a place where audio and video of a person describing their mental health may pass to a party the customer has not been told about, and the company names none of them.

The intellectual lineage is described where the technical stack is not. Public material traces the approach to roughly two decades of brain science research associated with the founding scientist, which explains where the ideas came from and says nothing about what runs in production.

The training question compounds it. Whether interviews recorded in a school, a workplace or a correctional facility feed model development is unaddressed, and the settings involved include populations whose consent is most complicated. The anonymity of collection does not answer it: a recording with no name attached is still a recording of a face and a voice, and those are the inputs the models read.

One real peer reviewed study, correctly disclosed in the journal and framed differently outside it, with the vendor's own link still pointing at the wrong version.

The study itself is solid, and it is not a preprint. It appeared in the Journal of Affective Disorders in 2024, a respected psychiatry title, led from the University of Texas at Austin with the Georgetown University Medical Center and the University of Arizona. Nearly 400 participants aged 18 to 74 completed both a bot administered interview and a gold standard depression questionnaire, and the model performed comparably. A subgroup analysis reporting no evidence of bias by gender, age or race accompanies it.

The framing around it diverges from what the paper says about itself. Both press coverage and the company's own release present a rigorous University of Texas study demonstrating clinical validation. The paper's declarations state it was funded by Aiberry, that two authors are Aiberry employees, and that a third is an Aiberry research coordinator. The employed author carries conceptualization, methodology, formal analysis and supervision, which is the methodological spine of the work. The journal disclosure is exactly right; the independent university validation story built on top of it is not.

The citation is also wrong. The research page's Read the Study control links a PsyArXiv preprint at psyarxiv.com/btz6n rather than the published article, so a reader following the vendor's own link reaches the unreviewed version of work that has since cleared review. The stronger artifact exists and the site points past it.

Scope is one condition, and the gap has widened. Depression is validated. The products now market twelve screenings covering anxiety, drug and alcohol use and eating disorders, and the voice assessment is validated for none of them. The screening instruments themselves carry their own established validation, which is a different thing from this company's model being validated to score them.

Commercial evidence is close to absent. No customer count, facility count or named reference outcome appears in any of the four marketed settings. The news feed stops in 2023. The figures published instead are satisfaction: 88 percent of users would use it monthly, two of three believe it gave better information. Those measure willingness, not outcome.

The product makes one architectural minimization decision and one real disclosure about what organizations receive, and describes nothing about how anything is actually handled.

The minimization is a design choice rather than a policy sentence, and it now covers both named products. Digital MindCare and Smart AI MindCare are both stated to require no sign in and to be completely anonymous. In schools, workplaces, correctional facilities and community organizations, which is where volume and reluctance are both highest, removing the identifier reduces the consequence of a breach rather than promising to guard against one.

The second disclosure is about distribution rather than storage: non clinical organizations receive only aggregated, de identified insights and never individual level responses, and the company states it shares no identifying information with the organization at all. That answers the question a student or employee would actually ask.

Handling detail is absent. No encryption statement, no key management, no retention schedule, no access control model, no audit logging description and no deletion process is published for either product. What exists is the phrase that data is protected using industry standard safeguards, which is an assurance rather than an answer.

The training question is unaddressed in both directions. Nothing states whether recorded interviews contribute to model development, under what consent, or whether an organization can decline. The material in the voice product is audio and video of a person describing their mental state, which cannot be de identified the way structured data can, because the face and the voice are the identifier and are simultaneously the input the models read. Anonymous at the point of collection and absent from the training corpus are different properties, and only the first is claimed.

The settings sharpen it. Recordings of incarcerated people and of students discussing their mental health are unusually consequential if retained.

A self asserted compliance claim with nothing external behind it, alongside one real statement about what organizations receive.

HIPAA compliance is represented by a badge in the site footer, with no attestation, no assessor, no scope statement and no external verification of any kind. HIPAA has no certifying body, so a badge is a claim about the company's own view of itself.

What sits beside it is more useful than the badge. The company states that it shares no identifying information with the deploying organization, that both named products require no sign in, and that non clinical organizations receive only aggregated, de identified insights and never individual level responses. For a school or an employer, that answers the question a participant would actually ask, and it is a structural answer rather than a promise.

Business associate agreement handling is not addressed anywhere. No template, execution requirement, statement of which entity contracts, or subcontractor flow down position is published. That matters most for the quoted enterprise tier, where clinical solutions are offered and an identified record is implied, and it is the tier with the least published detail.

The site links a single privacy policy from its footer. An earlier structure splitting policies by product, one for anonymous screening and one for the account based clinical product, was not re-verified on this reading, so whether the split survives the current site is open.

One regime specific to the named settings is unaddressed. Behavioral health and recovery centers handle substance use disorder treatment records under a separate federal confidentiality regime with its own consent and redisclosure requirements, and a recovery center is one of the four settings marketed, with drug and alcohol screening among the instruments offered.

A single footer badge is the entire published security posture, and it represents the one claim in this area that no external party can attest.

No security page, trust center, external attestation, penetration testing statement, vulnerability disclosure policy, encryption description or compliance documentation available under agreement is published anywhere on the site. The footer carries a HIPAA badge and nothing accompanies it. HIPAA has no certifying body, so the only security signal published is one that certifies nothing. The customer facing answer on data protection reads that data is protected using industry standard safeguards, which is an assertion rather than an examination.

An external attestation is increasingly an entry requirement rather than a differentiator, so its absence will surface in the first security review at any health system, correctional agency or university.

The data at issue makes the gap more serious. The voice tier records audio and video of people describing their mental state, including incarcerated people and students. That material is unusually identifiable, unusually sensitive and unusually damaging if exposed, and nothing published indicates that anyone outside the company has examined the controls protecting it. Anonymity at the point of collection reduces what a breach would reveal about identity and does not reduce what it would reveal about content.

A scope position is now stated and no device determination is, which leaves the classification question open rather than answered.

Some of what is absent counts in the company's favor. It makes no regulatory claim it has not earned. There is no registration badge presented as a quality mark, no designation described as a clearance, no approval language, and that silence is the honest position.

The company now states a position on what the product is for. It says plainly that it does not replace counseling or therapy, describes itself as an early supportive touchpoint helping people understand what they are experiencing and guiding them toward appropriate resources, and publishes routing to support and expedited crisis referral as the mechanism. That is a scope statement, and it is the kind that keeps a screening tool on the support side of the line.

What is still absent is the determination itself. No device classification, no classification rationale, no clearance and no statement that the product is positioned outside device regulation appears anywhere.

The question is live rather than academic, and the product structure sharpens it. Both named products are anonymous with no sign in, so no clinician stands between the model and the person receiving a result by default. The Smart AI tier publishes a full transcript alongside its output, which gives it an unusually strong reviewability argument wherever a clinician is involved; in an anonymous deployment there is nobody to review it. The entry tier delivers established screening instruments, which sit in a long understood regulatory position, while the voice assessment that scores a person from how they speak does not.

Two settings carry their own regimes that nothing addresses: screening in correctional facilities and screening students in higher education each raise consent and mandated reporting questions distinct from device law, and both are marketed.

A subgroup analysis is published in a peer reviewed journal, and it carries a statistical caveat the company does not state and a new language gap it has not addressed.

The substance is real. The 2024 study reports no evidence of bias by gender, age or race, and the company describes its training and validation population as reflecting the composition of the United States population, spanning ages 18 to 74 and depression severity from minimal to severe. The positioning is specific rather than general: the company contrasts itself with platforms relying on vocal biomarkers alone, which it identifies as commonly showing gender bias, and offers multimodality as the mitigation. That is a testable claim about a named competing approach.

The caveat is what the finding can carry. A null result in roughly 400 participants is an underpowered null, not a demonstration of fairness. Divided across racial categories, sexes and age bands, that sample can detect only large disparities; moderate ones would be invisible. Absence of evidence of bias and evidence of absence of bias are different findings, and the marketing treats them as one.

The company's own research page shows the slippage. Of its four bias cards, only the gender card reports a finding. The age card states a validated range. The ethnicity card states that participants reflect the United States population, which is a claim about who was recruited rather than a result about how the model performed on them. Representativeness of a sample is not subgroup performance, and the page substitutes one for the other.

Three gaps remain and one is new. No subgroup discrimination figures are published, only the null. Skin tone is not analyzed, which is the variable that drives differential error in facial analysis and is not interchangeable with self reported race. Primary language is not examined at all, and the product is now offered in Spanish while the validation study was conducted in English, so a second language is being sold against evidence from the first.

A recruitment claim also appears that the study does not support: pipelines described as designed to eliminate bias including by sexual orientation, a category the published analysis does not cover.

Responsibility is unallocated, and the product architecture makes the question harder here than for a conventional clinical tool.

No service level agreement, accuracy warranty, performance guarantee, indemnity, remediation commitment or published statement of recourse exists. Terms of use sit behind the application login at a separate domain rather than on the public site, so the contractual position cannot be examined before contact.

The published evidence does not fill the gap. Clinical equivalence to a questionnaire is a research finding about average behavior across a sample, not a commitment about an individual result, and because no operating threshold is published there is no stated false negative rate for any score.

What sets this company apart is who absorbs a miss. Both named products are anonymous with no sign in, so in the ordinary deployment there is no clinician, no identity and no account. A missed indication of risk reaches a person with nobody positioned to notice, and the deploying school, employer or facility has no way to know it happened, since non clinical organizations receive only aggregated data by design. The company has published a mechanism that acknowledges this shape, routing to support and expedited crisis referral, and has published nothing about what happens when the mechanism does not fire.

The entry tier carries a different exposure from the voice tier and the same silence covers both. Delivering an established screening instrument and scoring a person from how they speak are different acts with different failure modes, and one contract appears to cover both.

Not one record system is named, and the absence of integration is now presented as a feature rather than explained.

No electronic health record integration, interface standard, marketplace or partner listing, application programming interface documentation or data exchange description is published. The company states that getting started requires no complex integrations or training, with a customized link configured within a few business days, which confirms the position rather than answering it. The one integration claim previously made, that the corporate wellness version connects easily with leading corporate wellness platforms, names none of them and covers the setting where a clinical record system is least likely to be involved.

The absence weighs more heavily here than it would for a wellness product, because two of the four marketed settings are clinical. Behavioral health and recovery centers run on record systems, and a screening result that cannot reach the chart produces an outcome measure living in a separate dashboard, transcribed by hand to become part of the clinical record. The company's own description reinforces it: organizations receive an analytics dashboard, which is a destination rather than a connection, and non clinical organizations receive only aggregated data, which cannot reach a chart by design.

The anonymous architecture is part of the reason, so the gap is not simple neglect. A result with no identity attached has nowhere in a patient record to go. That is a coherent design choice, and it means a buyer wanting screening results in the chart needs the quoted enterprise tier, where clinical solutions and custom integration are listed without detail.

A correctional or university deployment raises the same question against different systems, and neither is addressed.

The delivery model can be inferred and is never stated.

Only the product's visible shape gives it away: a browser based application served from a company subdomain, reached through a customized link configured per organization, with a separate login for the enterprise application. That is a multi tenant cloud product by every visible sign, and the company nowhere says so.

Everything a buyer would need is absent. No cloud provider, hosting region, data residency commitment, tenancy model, segregation description, backup or disaster recovery position, or customer controlled deployment option is published.

Two of the marketed settings make this more consequential than a bare omission usually is. Correctional agencies are government buyers who ordinarily impose their own hosting, authorization and data location requirements, and public universities frequently carry state level data residency obligations. Nothing indicates whether either can be accommodated, and a procurement team in those settings will hit this in the first security review rather than the last. The quoted enterprise tier lists custom implementation and integration without saying whether deployment isolation is among the things that can be customized.

Anonymity does not remove the question. A screening without an account still generates responses, and in the voice tier audio and video, that must be transmitted and processed somewhere. Where that happens, and what is retained after a score is produced, is unstated.

One tier now carries a figure, which is enough to size the entry cost before making contact.

Digital MindCare is published at $5,000 a year, described as simple pricing with no contracts and no usage limits, and the inclusions are listed: twelve digital mental health screenings, analytics and reporting, routing to appropriate support, expedited crisis referral, and Spanish availability. An annual organizational license with no usage cap is a stated basis of charge, and for a screening product deployed across a school, a workplace or a facility the absence of a per screening meter is the fact that decides affordability.

Smart AI MindCare, the voice and visual assessment tier, is quoted rather than priced, listed as adding the voice screening experience, advanced analytics, a custom URL and five demographic questions on top of the entry tier. A custom enterprise tier covering clinical solutions, integration and workflow automation is also quoted.

So the base cost and what the next tier adds are both published, and what the next tier costs is not. There is also no self serve route: every commercial path terminates at a demo booking form.

Four buyer settings compound the remaining uncertainty. A correctional facility, a university counseling service, a corporate wellness administrator and a recovery center have different budget structures, procurement routes and volume profiles, and nothing indicates whether the annual license holds across all four or whether population size moves it.

No return proxy is published. The commercial argument is early engagement against a provider shortage, and no cost per screening, time released figure or throughput number supports it.

Breadth of setting and of screening menu is real, and breadth of validated clinical scope is not.

Four settings are named, and each is a distinct deployment context: corrections, corporate wellness, behavioral health and recovery, and higher education. Reaching into correctional facilities and universities is unusual and addresses populations conventional behavioral health vendors do not target. The anonymous, login free design is well matched to the school and workplace contexts where an identified health record would deter participation.

The screening menu has widened. Twelve screenings are listed covering depression, anxiety, drug and alcohol use and eating disorders, and Spanish is now offered on the entry tier.

Validated scope for the company's own model is one condition. The published study covers depression in adults aged 18 to 74, in English. The established instruments delivered through the platform carry their own validation, which is a separate thing from this vendor's voice model being validated to assess those conditions, and nothing published distinguishes the two. Spanish availability is offered against evidence generated in English, which matters for a model that reads word choice and speech patterns.

Two populations sit inside the named settings and outside the validated range. Higher education includes students under 18 in some programs, and the study population starts at 18. Correctional populations carry consent and coercion considerations that nothing published addresses, and a screening instrument administered where participation may not be freely refused is a materially different product from the same instrument in a clinic.

The screening count is inconsistent: it is published as thirteen on the pricing card and twelve on the product pages.

Nothing addresses markets outside the United States.