This Startup Catches Fraud at Scale

Karim describes Variant as a company building AI agents for content, fraud, and identity reviews. It is announcing a $21 million Series A after three years in stealth. Karim says the work involves sensitive data and that publicizing some use cases could enable more fraud or abuse.

Karim explains how Variant helps GoFundMe check whether fundraisers are legitimate and whether donations could create compliance risks. After a major event, fraudulent fundraisers may claim to represent victims or their families. Variant’s agents use account history, identity details, fundraiser content, and GoFundMe’s policies to assess cases that human analysts previously reviewed.

Variant combines customer procedures with internal and external data, including more than 100 business registries and information from the open web. Karim says customer information can be scattered across five to ten systems, sometimes behind interfaces built for human reviewers. He argues that agents can combine signals and adapt more quickly than a patchwork of rules, classifiers, and manual reviews.

Karim and co-founder Michael met on Apple’s fraud engineering team and started Variant to make those systems more efficient and adaptable. Their first customer, IAC, used the product to review marketing content against compliance guidelines that were difficult to encode in traditional classifiers. Karim also recounts a crash that left him with a broken spine and leg, and how the team continued while he recovered.

Karim says Variant had 12 employees, including five software engineers, and was hiring across engineering. Agents handled about 99% of cases in one workflow, while the most complex cases still needed human review and a strong investigative dashboard. Engineers used coding agents to increase output, and a nontechnical customer success manager could use them to ship some simple customer-requested features.