Pixel Systems

Stockholm-based Pixel Systems builds software that checks whether images, video and documents are genuine. Its product looks for signs of AI generation and manipulation, so that a person deciding on an insurance claim, an invoice or a piece of evidence has something firmer than a visual impression to go on.

One pass over a suspect file

The company says a single check surfaces edited areas, cloned regions, tampered metadata and the fingerprints left by generative tools. It describes multi-signal verification that covers content from consumer AI apps as well as more advanced generative models. A result comes back as a verdict, the signals behind it and a record meant to support a decision that may be challenged later. On its site the result view lists dimensions such as AI detection, reverse image search and file history, each with a score and a confidence level.

Fraud cases it points at

Pixel Systems pages are organised by the situations where doctored files cost money. In insurance the aim is to catch fabricated claim evidence. For finance and audit teams it looks at altered receipts, invoices and scans. Legal and defence users are offered help triaging digital evidence, media organisations can screen submissions, and platforms and marketplaces can route risky uploads to trust and safety staff. E-commerce sellers are shown how to flag fake refund requests and seller listings.

API, SDKs and a no-code dashboard

Teams can connect through a REST API with reference documentation and TypeScript SDKs; jobs are fetched by ID from api.pixelsystems.com with bearer-token authentication. Those who prefer not to write code can upload files to a web dashboard and read risk scores there, with sections for jobs, API keys, insights and activity.

Independence and security stance

The company states that it does not build AI generators, which it presents as the reason it has no stake in the outcome of a check. It cites GDPR with EU hosting, and ISO 27001 and SOC 2 controls covering encryption, tenant isolation and least-privilege access. It says its models are proprietary and based on current media forensics research.

Key facts

Headquarters: Norrlandsgatan 22, 111 43 Stockholm, Sweden Funding: $3.1 million seed round, as announced on its site Team background: founding team with academic roots at KTH Royal Institute of Technology, plus people from finance, infrastructure and high-scale software Hiring: engineers, ML researchers and enterprise go-to-market roles in Stockholm Security: ISO 27001, SOC 2, GDPR