Samsung and Reality Defender to run deepfake detection directly on Galaxy devices

Samsung Electronics Singapore and Reality Defender have signed a memorandum of understanding to build deepfake detection that runs directly on Galaxy hardware, analyzing AI-generated voice, video and images on the phone itself rather than sending them to the cloud. According to a report on dev.to, the agreement was signed during the GovWare 2026 security conference and its first customers will be enterprises in Singapore.

The fraud numbers behind the deal

The figures Samsung presented alongside the announcement, as relayed by dev.to, explain the urgency. A 2024 survey by Regula found that 56% of Singapore businesses faced fraud attempts involving fake audio, while 52% dealt with manipulated video. The measurable damage is larger: victims of scams impersonating public officials in Singapore lost at least S$242.9 million — roughly US$188 million — during 2025, accounting for 26.6% of everything stolen through scams in the country that year.

Ben Colman, co-founder and CEO of Reality Defender, framed the problem plainly in the announcement: for many organizations, the first place a deepfake lands is an employee's phone. A cloned voice built from a few seconds of sample audio is enough to request an urgent transfer, and the first device receiving that request is rarely a workstation monitored by an IT department.

Why analyze on the device

Privacy is the central design constraint. Routing every suspicious call or message to an external server for review exposes sensitive conversations and adds waiting time at the exact moment someone must decide whether to follow an instruction. Running the check inside the phone shortens that path: sensitive material never leaves the device just to find out whether it is fake.

Knox Vault and per-app encryption

The technical proposal pairs Reality Defender's detection platform with Samsung Knox, the security architecture that protects Galaxy devices from chip level to applications. Two components carry the design. Knox Vault keeps the most sensitive data in an isolated environment, separated from the main operating system. Knox Enhanced Encrypted Protection, known as KEEP, creates per-application encrypted spaces so each program can access only its own protected information.

Reality Defender contributes the detector itself. Its platform identifies AI-generated voice, video and images in real time within the workflows where businesses and governments make decisions, and it works statelessly — in Colman's words, nothing is recorded and nothing is stored. The companies do not plan to build a library of employee voices or faces for later comparison; analysis happens on the content received and is then discarded.

Timothy Tan, director and head of integrated business at Samsung Singapore, said the goal is to develop practical on-device detection that flags threats before data needs to be processed in the cloud. The same capability could support know-your-customer identity verification, where a fabricated video can open accounts or authorize transactions in someone else's name.

A deliberately narrow first rollout

The initial scope is limited. According to dev.to, Samsung confirmed the solution will deploy on the Galaxy S26 series and later models, plus high-end Galaxy tablets, with Singapore enterprises as the target audience. There is no confirmed launch date and no public pricing: the memorandum covers development and deployment, not a product already available in stores.

For ordinary Galaxy owners, nothing changes yet. SamMobile, a publication specialized in the brand, cautioned that the initiative is enterprise-focused and that no equivalent plans for consumer phones are known — though several Knox capabilities have historically started as corporate tools before reaching a wider audience.

Why it matters

Moving deepfake detection onto the phone removes the cloud from the verification loop, which addresses both the privacy cost of uploading private conversations and the delay that comes with it. If the approach works, it offers a template for AI content verification that enterprises could adopt without renegotiating how their data is handled. The unanswered question is performance: Samsung and Reality Defender published no accuracy rates or independent test results, and that number will decide whether the tool actually stops fraud or merely adds alerts that employees learn to ignore.