SESAMm

The removal test leaves an unreadable archive. Natural language processing is the entire mechanism by which 20 billion articles and messages in dozens of languages become controversy flags, sentiment scores and investment signals mapped to five million named companies, and the stack has moved with the field, with the company now describing real time reputational and sustainability risk signals produced using large language models and generative techniques. A separate quantitative platform turns those outputs into systematic signals. No rules based approach reads that corpus.

The primary output is an alert or a score that an analyst acts on, and the design assumes human response, with customers described as proactively responding the instant an alert arrives about a supply chain partner or portfolio company. Delivery into email and deal management systems reinforces that a person is the recipient.

The qualification is the quantitative side, where signals feed systematic strategies and are consumed by models rather than read, and nothing describes confidence handling, false positive review or what a customer should do with a controversy flag before acting on it.

One accuracy claim is published and it needs stating precisely: after fine tuning with a ratings partner's expertise, the chief executive states the platform reaches 100 percent accuracy on the detection of critical controversies. That is scoped rather than impossible, since near complete recall on a narrow, well defined class of high severity events is achievable, and it differs in kind from claims that contradict what is knowable.

It is nonetheless unfalsifiable as published, with no denominator, no methodology, no false positive rate and no independent validation, and a perfect score asserted without any of those is weaker evidence than a lower one properly measured. Nothing else on validation was located.

One of the world's largest private equity firms is named as both investor and client, and its chief digital officer is quoted saying the analytics have supported deal sourcing, diligence and portfolio company value creation over more than two years.

Three further customers are named across three continents: a major French regional bank that selected the platform in December 2025, a South Korean asset manager building machine learning models on the alternative data, and a European ratings specialist that launched its own small company monitoring product on top of it. The venture arm of a large European bank co led the 35 million euro Series B2, taking total funding above 50 million. The team of more than 50 spans six offices across France, the United States, the United Kingdom, Japan and Tunisia, and the company holds two industry awards.

No data boundary statement was located, and the exposure is the one recorded at other deal intelligence platforms: what a customer searches reveals its intentions. A private equity firm running diligence on an acquisition target through this platform discloses that interest, and the platform serves competing firms in the same market, one of which is also a shareholder. Nothing states whether queries are retained, whether search activity informs anything, or how one client's research is separated from another's.

Structurally favourable because the subjects are companies rather than consumers, with the corpus drawn from public articles, blogs and messages and the output mapped to corporate entities, so no personal financial data is processed. Two qualifications keep it from higher. Controversy detection frequently concerns the conduct of named individuals at those companies, and the source material includes social messages written by identifiable people. Held at B also because no data processing terms, retention schedule or subprocessor list was located across a six country footprint including two European jurisdictions.

No attestation, certification, trust centre or enumerated framework was located. A major private equity firm, a large European bank's venture arm, a French regional bank and an Asian asset manager have all completed supplier assessment, so the underlying work exists, and for a platform whose customers reveal live acquisition interest through their queries, published controls around that search activity would be the natural disclosure.

No supervisor, statute or instrument is named, which is a specific gap for this product rather than a general one. European sustainability disclosure rules determine what asset managers must report about the sustainability characteristics of their holdings, and controversy data feeds exactly those assessments, while providers of sustainability ratings have themselves recently been brought under supervision in the European Union. The company partners with a regulated ratings specialist and names neither framework.

No individual is assessed and the adapted exposure is consequential and, on the company's own logic, self undermining. A controversy flag affects a company's access to capital, so a false positive damages a business that did nothing wrong and never learns why.

More fundamentally, detection depends on media coverage, so a company operating in a well reported market with an active press generates more detectable controversies than an identically behaving company in an under reported one, which means the method can penalise transparency and reward opacity. That sits awkwardly beside the company's correct argument that opacity around smaller businesses distorts investment. No coverage or false positive analysis by market, language or company size was located.

No guarantee, indemnity or falsifiable commitment was located. The investing customer can test signals against outcomes over time and has the ratings partner relationship as a quality check. The company being assessed has nothing at all, and its position is the weaker one: it is not the customer, is generally unaware it is monitored, cannot see what controversy has been attributed to it, and may find capital harder to raise because an automated system read coverage about it in a way no person reviewed.

The input side is quantified and characterised with unusual precision for a data business, at more than 20 billion web articles and messages growing 20 percent annually across dozens of languages, drawn from public sources including news, blogs and message content, which tells a buyer the scale, freshness and nature of the evidence behind any signal. The language engine is the company's own and the model class is disclosed as including large language and generative models. What is not named is any individual source, aggregator or model provider, and no subprocessor list or hosting arrangement was located.

Four delivery routes are described and one names the system that matters for the core buyer, with insights delivered into the deal management platform private equity firms already run their pipelines on, so a controversy signal appears against the target in the workflow rather than in a separate tool.

The interface option brings the language engine into a customer's own systems, a hosted dashboard provides analysis, visualisation and notifications, and alerts route to email or onward systems including customer relationship platforms. No market data, portfolio or risk system integration is named.

No hosting provider, region selection, residency commitment or private deployment option was located. Operations span six offices across three continents with customers in Europe, North America and Asia, several of them regulated institutions whose own supervisors expect documented arrangements for where outsourced analytical processing occurs.

No pricing, packaging or basis of charge was located. Several consumption routes are described, spanning an interface for embedding the engine in a customer's own systems, a hosted dashboard, alerting and integration into third party deal software, and those would ordinarily be priced differently. Nothing indicates whether charge scales with companies monitored, queries, users or data volume.

Six buyer types are served, spanning private equity firms, hedge funds, asset managers, banks, ratings agencies and corporates, with the ratings agency relationship notable because it means a firm whose own business is assessment builds on this engine.

Subject coverage is the standout at five million public and private companies, and the private half is the point, since the company argues that traditional providers and rating agencies lack coverage of smaller businesses and that the resulting opacity distorts investment decisions. Language coverage spans dozens, and offices sit on three continents.