S

Securiti

AI appears here in three ways, and none makes the models the product a legal buyer pays for. First, as subject matter: the AI Governance, EU AI Act and NIST AI RMF pages sell the ability to govern a customer's AI, which is managing someone else's models. Second, as infrastructure for the customer: Gencore AI prepares data for training, runs prompt, retrieval and response firewalls, and builds copilots, which is AI plumbing rather than AI doing the buyer's work.

Third, as technique inside classification and the DataAI Command Graph. Remove the models and most of the platform remains saleable: data discovery and scanning across a stated thousand or more integrations, a catalog, lineage, a data map producing records of processing activity, data subject request workflow, consent capture, assessment templates and breach notification.

Accuracy is asserted repeatedly in the marketing and measured nowhere. Product copy claims accurate data classification, discovery of shadow and cloud native assets, and unified intelligence. Classification precision is the key quality measure for a discovery product: a missed store of personal data is the failure that matters, and over broad matches create work that erodes trust in the tool. No precision or recall figure, benchmark, test set or false positive rate is published on any page read.

The platform does not retrieve legal authority or produce citations a reader would open. The agreement is more candid than the marketing: Securiti warrants substantial conformity with the Documentation, not the accuracy of any classification result.

Oversight is sold as a product, not described as a control over the vendor's own output. Securiti ships real runtime controls: prompt, retrieval and response firewalls for large language models, and Agent Commander for detecting and undoing AI agent mistakes. All of them govern the customer's AI systems rather than Securiti's classification and assessment engine. On its own side, nothing says what runs unattended, what confidence threshold makes the system defer, where a reviewer sits relative to a classification decision, or what happens when a classification is wrong.

Assessment Automation implies people write assessments, without describing a check on machine output. The agreement is the only place a supervision expectation appears, and it points at the customer: the product augments rather than replaces professional advisors, and the customer must confer with legal counsel as needed. That assigns responsibility rather than describing a mechanism.

Named organizations and individuals appear with titles and dates, but with no figures for what changed. The Spotlight Talks series has on the record interviews with a director of global analytics at Dye and Durham, an SVP of product at Walker and Dunlop discussing a 135 billion dollar portfolio, and named executives at Sanofi, Volkswagen and International Flavors and Fragrances, dated April to June 2025. That is interview material, not outcome measurement: no before and after metric, deployment scale or time saving is attached.

Analyst recognition is extensive but is not deployment evidence: GigaOm, Frost and Sullivan, IDC MarketScape, Forrester Wave, Gartner Cool Vendor and an RSA Conference Innovation Sandbox win. Section 9.2 of the customer agreement requires customers to join Securiti's reference program and develop a profile with an executive quote and logo, so testimonials are a contract term rather than purely voluntary.

The architecture is documented in detail, and the data use question is answered in the vendor's favor. In the agreement or its security exhibit: a virtual database instance for each customer separates its data logically and is destroyed when the customer leaves. Personal data the platform identifies is put through a one way irreversible hash, with a commitment that personal data is never captured in clear text in logs or databases.

Confidentiality is mutual, with notice and a chance to contest compelled disclosure. The customer owns Customer Data, and all of it is deleted from online systems within one business week of a confirmed request. Against that, section 2.1 grants Securiti a royalty free, worldwide license to use Customer Data both to provide the service and for enhancing product or services, with no carve out for model training and no definition of enhancement. Privilege and work product are not addressed, and there is no position on third party model providers.

A real position on advice versus tooling is published in the agreement itself. The disclaimer at section 5.4 says the customer acknowledges the product is meant only to augment its privacy practices, not to replace legal and other professional advisors, and that the customer is a data controller responsible for the data it collects and its own privacy policies. Section 2.5 adds that the customer takes full responsibility as controller, warrants it has met transparency obligations and obtained consents and legal bases, and must confer as needed with legal counsel to confirm and maintain compliance.

That allocates professional judgment to the person relying on the output. No jurisdiction limit is named for the tool's coverage, nothing addresses the supervision or competence duties of the practitioner using it, and no professional guidance is referenced.

Nothing published addresses governance of Securiti's own models, which contrasts with what the company sells. Securiti publishes extensive material on AI governance frameworks as product capability, with pages for the EU AI Act, the NIST AI Risk Management Framework, the OWASP Top 10 for LLM Applications and CDMC, and sells an AI Governance module. None of that is Securiti's own governance. No responsible AI page, AI policy, ethics statement, AI governance committee, named accountable owner, testing regime before release or bias evaluation is published.

The site navigation was read in full and has no such page. The security exhibit names a security official and a cross functional Security Council, which is information security governance, not AI governance. A vendor selling AI governance publishes none of its own.

The security exhibit to the customer agreement is specific across most of the ground; a named subprocessor list is missing. Deletion has a stated window: on a request by ticket or email, Securiti deletes all Customer Data from online systems within one business week of confirmation, keeping data only where law requires. Incidents have a deadline and method: notice of a security breach as soon as practicable and no later than seventy two hours after Securiti becomes aware, by email with a read receipt to a designated address.

Securiti may not tell third parties without approval, and the customer alone decides whether affected consumers are notified. Access control is role based, reviewed regularly and monitored, with only a stated subset of staff able to reach customer data. Testing has a cadence: annual third party penetration tests and audits, weekly internal scans, and disaster recovery tested twice a year, with a summary available on request.

Devices carry at least AES-128 full disk encryption. Third parties appear only as categories such as IT service providers and cloud providers, and the CCPA addendum has the customer approve transfers to unnamed affiliates, providers and vendors.

The agreement covers indemnity, a cap, a warranty and insurance in specific terms. The indemnity covers more than intellectual property. Securiti defends the customer against third party claims that use of the product infringes a US patent, copyright, trade secret or trademark, and separately against claims arising from Securiti's use or disclosure of Customer Data in breach of the agreement. Direct damages are capped at the lesser of amounts paid under the order form in the previous twelve months or one million dollars, with consequential damages excluded for both sides and the customer's payment obligations carved out.

A warranty commits the product to substantially meet the order form, substantially conform to the documentation and be free of malicious code, with performance again, termination and a pro rata refund as the exclusive remedy. An insurance position is published: errors and omissions, professional liability and cyber cover of at least three million dollars per claim and in aggregate, held through the term and two years after, with thirty days' notice of cancellation.

A service level agreement adds 99.5 percent availability, tiered credits and a termination right below 92 percent for three months running. Nothing is specific to wrong AI output.

Real integrations exist, are named, and stop short of documented depth. Securiti states more than a thousand integrations across data systems and names the principal ones on its own navigation: AWS, Google Cloud, Azure, Snowflake and Databricks each carry a dedicated page, with a connectors index behind them. What travels is describable at a high level from the product set, since discovery, classification, access intelligence and lineage all operate by reaching into those systems and reading their contents, and downloadable components are deployed inside the customer's own environment for parts of the platform.

What is missing is the implementer's view: no API reference or connector specification is published on the pages read, and nothing describes the direction or granularity of what moves for any individual system. A documentation site exists at docs.securiti.ai, listed among the company's own internet-facing assets in the published system description, and was not opened, so depth was neither confirmed nor excluded.

The deployment model is stated plainly, with partial residency detail. The published system description sets out a multi tenant cloud service on AWS and GCP, with a virtual database instance for each customer providing logical separation, plus downloadable components that run inside the customer's own environment. Residency is real: instances sit in several geographically distributed data centers, and each instance serves customers from a specific geography as a standalone offering with no data exchange between instances, which answers the processing question directly.

Two clouds are identified by their endpoints: a Global Production Cloud and an EU Production Cloud at app.eu.securiti.ai with its own status page. Resilience detail is published: daily backups copied to a data center in another region, a pilot light disaster recovery strategy, failover across availability zones, and a stated 24 hour RTO and RPO. The full list of regions is not given beyond the EU and Global clouds, and nothing describes what changes between deployment options.

The certifications are real and named, and none of the evidence can be reached without asking. Securiti states SOC 2 Type II certification, with the report available on request to prospective and current customers, and holds ISO 27001:2022 and ISO 27701:2019, the privacy information management standard. The agreement backs the SOC 2 position, committing Securiti to provide its latest SOC 2 report or an industry standard successor on request.

The published system description names AWS and GCP and the shared responsibility split, annual third party penetration tests and audits, weekly internal scans, and a disaster recovery test summary on request. There is no trust center and nothing to download. No auditor or certification body is named for any of the three, and no certificate date, examination period or scope appears. Every route to the evidence is a request.

The platform is built around models and names none of them. Securiti describes a knowledge graph at the core, classification across structured and unstructured data, preparing data for vector databases, curating and cleaning data for model training and tuning, and prompt, retrieval and response firewalls for large language models. All of that implies models, and no model, model family, provider or architecture is named on any page read.

Nothing says whether a third party foundation model is called in classification or in the copilot products, and nothing rules one out. AWS and GCP are named as infrastructure, which says where workloads run, not whose models they are. No commitment to notify customers when the models change is published. The architecture is described in real terms; nothing underneath it is identified.

No figure or tier is published, but the charging unit can be read from the agreement. The site navigation was read in full and has no pricing page; every commercial route is a demo request or contact form. The customer agreement caps the number of Authorized Users at the figure in the order form, which makes named users the unit of charge. Fees are quoted and paid in US dollars. Payment obligations cannot be canceled, are not prorated for partial months and are not refundable.

Late payment carries interest of one and a half percent a month, and subscriptions renew automatically for terms equal to the first unless either party gives thirty days' written notice. The service level agreement adds credits of four, six and ten percent against availability levels. A buyer can see the shape of the commercial relationship in advance but learns nothing about cost: no rate, band, package or tier name is published.

Coverage is documented across industries, regulations and technology, and the boundary is left open. Six industries have dedicated pages: financial services, healthcare, telecom, retail, travel and hospitality, and manufacturing. Regulations are named, with pages for GDPR, California's CPRA, Brazil's LGPD, Canada's PIPEDA, China's PIPL, the EU AI Act, the NIST AI Risk Management Framework, the OWASP Top 10 for LLM Applications and CDMC, behind a wider list.

Technology coverage is put at more than a thousand integrations, with the major cloud and data platforms named. Nothing says where the product stops: no organization size, data volume, or environments and obligations it does not reach. The Roles navigation offers Data+AI Builders, Data Security, Data Privacy, Data Governance and Marketing, with no page for legal or counsel, although the privacy pillar is work a data protection officer or privacy counsel owns and the agreement tells the customer to confer with legal counsel.