Infinitus Systems

The agents are the product. They place and take the calls: benefit verification, prior authorization and claim status calls to payers, missing-information calls to provider offices, inbound copay and financial support calls from patients, health plan member services, health risk assessments, and medication adherence check-ins.

Lens scores both agent-led and human-led conversations, the copilots hand staff a live representative after the agent has worked the phone menu and the hold, and Studio lets customers build their own agents. Clinical Escalations runs alongside every call as a separate risk model. With the models removed, what remains is a portal, an API and a telephone line with nobody on it.

The oversight structure is unusually well described, and the company also claims its system does not miss. The structure: Clinical Escalations scores every turn of a patient conversation against the Schmidt-Thompson Call Prioritization Index and, when a threshold is crossed, fires a warm transfer, a 911 handoff or a nurse queue mapped to the customer's protocols; every exchange is logged for human review; for dosing information, contraindication disclosures and boxed warnings, the Agent Response Control system plays approved text verbatim with no language model in the loop; and a clinical team of doctors and nurses evaluates every agent conversation.

The claim: the product page says the system has 'achieved a 0% under-triage rate. Meaning no emergency is missed', the clinical evidence page promises 'No clinical signal goes unaddressed', and verbatim control is said to deliver 'complete SOP compliance'. A buyer who believes no emergency is missed will not build the review step that would catch the one that is, and review of a logged conversation shows what was escalated, not what should have been. The 0% figure is published without the number of calls or emergencies behind it.

The approach is named in useful detail; the models and their versions are not. The company describes a patent-pending Agent Response Control system that removes the language model from the loop and plays approved text verbatim where a script must be followed exactly, controlled responses drawn from pre-approved knowledge bases elsewhere, a separate Clinical Escalations model that scores each conversational turn against the Schmidt-Thompson Call Prioritization Index, and training of that model on real patient conversations across literacy levels and communication styles.

Lens applies the same scoring to agent-led and human-led calls. What is not published is which language models sit underneath, how they are versioned, or when a customer is told the system has changed.

The host is named and the model layer is not. All services and data run on Google Cloud Platform in the United States. The company says it has business associate agreements with all of its downstream vendors, assesses each vendor's security, privacy and data handling at least annually, and requires them to hold a recognized certification such as ISO/IEC 27001 or SOC 2 Type II.

None of those vendors is named beyond Google, and no language, speech recognition, speech synthesis or telephony provider is identified, so a buyer cannot tell which parties hear a call or receive its transcript.

The company publishes its own research. Its clinical evidence page features a study of 6,000 real patient calls across three interaction types (health risk assessment, cholesterol education and PHQ-2 depression screening), measuring clinical accuracy, empathy and what the agents can handle, and reports agent scores of 5 out of 5 for empathy and 4.9 out of 5 for professionalism.

Named deployments include Amgen's patient access and reimbursement team, Accompany Health for triage and Zing Health, a Medicare Advantage plan, for member health risk assessments, with an anonymous Medicare Advantage case study reporting 90 percent positive member ratings.

The work is company run, with no independent evaluation, and it measures conversation quality and safety rather than the headline outcomes the home page leads with: three times faster time to therapy, up to five times return and more than 90 percent positive patient ratings, which carry no source, baseline or denominator.

Categorical commitments are published on the security page. Customer protected health information is processed only for the purposes specified in each customer contract, collected only as needed for the service, returned or securely destroyed at the end of the contract within a retention period set in the service agreement, and passed only to downstream vendors that have signed business associate agreements.

The privacy policy, which covers people who speak with the agents, adds that personal data may be used to improve the functionality of the services, which sits uneasily with the contract-purpose commitment. The exposure specific to this product is spoken disclosure: an agent reads patient identifiers aloud to payer representatives and provider office staff to complete a call, and the recording at the other end is outside the vendor's control.

Missing: whether conversations train any model (the Clinical Escalations model is said to be trained on real patient conversations), de-identification before any such use, and retention windows for call audio and transcripts.

Business associate status is stated on the company's security page: Infinitus follows the HIPAA Security and Privacy Rules, signs a business associate agreement with each customer whose protected health information it processes, abides by the requirements for business associates, and has business associate agreements with all of its downstream vendors. Staff are trained on HIPAA at onboarding and annually, and retention and destruction follow each customer's service agreement. The agreement itself is not published, so its terms and the tier it applies at cannot be read before contracting. The privacy policy covers people who speak with the agents.

SOC 2 Type II, named as the company's own, with the most recent report available to customers through their sales representative. The security page describes the program behind it in detail: annual external penetration testing, static, dynamic and software composition analysis, AES-256 encryption at rest with keys in Google Cloud Key Management Service, TLS 1.2 or higher in transit, intrusion detection, weekly log review, role-based access with supervisor approval, 24-hour deprovisioning, an annually tested incident response plan and business continuity plan, and sixteen named corporate security policies reviewed each year.

A Trust Center at trust.infinitus.ai is linked from the site footer; the report is still obtained by request. No HITRUST or ISO/IEC 27001 certification is claimed, and Google Cloud's certifications are the host's.

Clinical triage and medication management are marketed on patients' calls, and no regulatory position is stated anywhere the company publishes. The June 2026 launch of Clinical Escalations describes a system that 'ensures healthcare AI agents appropriately triage patients', reclassifying an administrative call as an emergent clinical event and firing a warm transfer, a 911 handoff or a nurse queue.

Its scoring follows a named nurse triage protocol, the Schmidt-Thompson Call Prioritization Index, so the logic can be checked against a nurse line's own. The product page says the system has 'achieved a 0% under-triage rate. Meaning no emergency is missed'. The December 2025 clinical agents are described as able to 'evaluate symptom severity and manage medications according to protocols set by a patient's care team'.

A nurse or 911 takes the calls the system flags as urgent; the agent alone decides the calls it scores as not urgent, and clinician review of the logs comes afterward. Separately, pharmaceutical program calls that surface adverse events carry the manufacturer's reporting duty.

A named evaluation process stands behind the agents. Every agent conversation is scored across professionalism, empathy, information accuracy, privacy, clinical safety and adherence to standard operating procedure, under a five-pillar framework overseen by a clinical team of doctors and nurses, with results fed back into the agents; the company's chief medical officer and medical lead are named.

Lens applies the same scoring to human-led calls, and a company study of 6,000 patient calls reports results on those measures. No bias or fairness evaluation is published, and no results are broken out by group. The company says its Spanish language agents carry 'the same rigorous evaluation standards as English', and that Clinical Escalations was tested across literacy, emotional state and clarity so it triages reliably however a patient describes their situation. Neither claim comes with figures by language or by group.

Mechanisms exist to challenge an output, and nothing stands behind it. Every exchange Clinical Escalations scores is logged so a clinical team can review each escalation decision, every agent conversation is evaluated by a clinical team against six measures, Lens scores both agent-led and human-led calls, and approved text for dosing and warnings is played verbatim. What is missing is anything the vendor answers for.

The 0 percent under-triage figure is published with no count of calls or emergencies, so it cannot be tested as an error rate, and no warranty, remedy or service credit attached to a wrong answer, a missed escalation or wrong benefit data written back to a customer's system is described. A patient whose emergency was scored below threshold has no route against the vendor described anywhere.

Record systems are named, but the integrations cannot be verified in a marketplace. The product page says the agents and copilots can be launched from within Epic, Cerner, athenahealth and Salesforce, and call results are returned as structured data to the provider, payer, hub, specialty pharmacy and manufacturer workflows that asked for them rather than left as a transcript.

Customers can also submit work and retrieve results through the Infinitus portal or API endpoints, and a Salesforce partnership covers Health Cloud, Life Sciences Cloud and Agentforce. No Epic Showroom or Salesforce AppExchange listing, integration documentation or named production deployment on a record system is published.

Hosting and residency are stated; the processing path is not. All services and data stores run on Google Cloud Platform in the United States, customer data is stored within US boundaries, backups are encrypted and kept in the cloud, and Google Cloud provides failover. Customer data is separated by unique customer identifiers with access controls, and development and production run in separate cloud networks. The company says it does not currently serve EU customers.

A single hosted option is offered. What is not published is where calls cross into other providers: the telephony carrier that places and records calls, and any speech or language model provider that processes audio or transcripts.

No price, rate card or unit of charge is published, and the route in is a demo request. The offer is packaged two ways, by buyer (pharmaceutical and life sciences manufacturers, payors and pharmacy benefit managers, provider organizations, and an AI first hub for patient support programs) and by product (AI agents, copilots, Lens conversation review and the Studio agent builder). The home page claims up to five times return, with no basis.

Its core work is insurance and access calls. The agents verify benefits and eligibility, follow up with payers on prior authorization and claim status, call provider offices for information missing from program enrollments, and check financial assistance and copay eligibility. Coverage now reaches well past that.

The company sells to pharmaceutical and life sciences manufacturers, payors and pharmacy benefit managers, provider organizations and patient support hubs, and its agents take inbound patient and member calls, run health risk assessments and medication adherence check-ins, and triage clinical risk on patient calls.

Validation stands behind part of it: Amgen, Accompany Health and Zing Health are named, the company reports use by eight of the ten largest pharmaceutical manufacturers, and its 6,000-call study covers health risk assessment, cholesterol education and depression screening. Most case studies are anonymous, and no specialty is named.