Reken launches on-device AI platform to detect phishing and impersonation scams

After two years of development, Reken has come out of stealth with an AI security platform that can analyze communications directly on users’ devices.

The Reken Private Core platform provides the underlying technology for Northstar, its application that can detect phishing, impersonation, business email compromise, and other social engineering attacks.

The Private Core is built around three components: the Model Zero, the Trust Sensor, and the Reken API that connects these capabilities to applications.

Northstar is currently being offered through an early access program.

Reken says the system has been tested in Fortune 500 environments but has not identified the participating organizations, according to a Fortune interview.

The company has not yet published performance benchmarks, customer case studies, or independent evaluations showing how accurately the system detects attacks in real-world conditions.

Reken was co-founded by Shuman Ghosemajumder and Rich Griffiths. Ghosemajumder held fraud and trust-and-safety roles at Google, where he helped build machine learning systems for detecting cybercriminals.

The company is now targeting a trust problem compounded by the use of generative AI. For instance, there is a growing problem with employees no longer being able to judge whether an email, message, or sender is genuine.

“We shouldn’t be forcing employees to become forensic digital investigators. We need just-in-time AI that detects the threats the human eye cannot see,” says Ghosemajumder in a  Fortune interview.

Reken is attempting to shift that decision away from employees and toward AI models operating at the point of communication receipt.

Private Core processes communications locally

The company says the Private Core uses small proprietary models that can run on ordinary devices without GPUs. This could give organizations greater control over confidential communications.

This architecture reduces the internet-exposed attack surface associated with cloud processing and avoids the usage costs when applications repeatedly query commercial foundation models.

The Model Zero performs the analysis of incoming threats. Reken says it combines proprietary AI models through ensemble learning to analyze data and detect attacks.

The Trust Sensor is the platform component that uses “advanced signals” to verify human activity and the data associated with outgoing communications and transactions.

The Reken API provides the integration layer. The company says any application can connect to the Private Core, while Northstar is described as operating within channels such as email, direct messages and group chats.

Reken wants to expand the deployment of the Private Core to the Reken Network. This will create a broader environment in which communication between participating users and organizations can convey trust signals.

Article Topics

AI fraud  |  cybersecurity  |  digital identity  |  Reken  |  startup