Prem launches CyberScan for Bitcoin security audits

Prem AI has launched CyberScan, an AI security audit tool for financial infrastructure that is initially focused on Bitcoin. It is designed to provide continuous code review between formal security audits.

CyberScan repeatedly analyses software repositories as code changes, returning prioritised findings with severity ratings and file-level references inside developers' existing workflows. Prem said the system preserves scan state if a review is interrupted, allowing it to resume without restarting.

The launch follows a major theft from Coldcard hardware wallets that exposed the risks of missed flaws in financial software. Prem said more than USD $116 million in bitcoin was taken from over 5,200 wallets after a firmware regression went undetected for five years.

According to Prem, the flaw was introduced in March 2021 when a build-time check meant to ensure the wallet used its hardware random number generator instead allowed it to fall back to a deterministic software pseudo-random number generator. As a result, wallet seeds became predictable, enabling an attacker to rebuild private keys offline without access to the physical devices.

Prem said the episode highlighted the limits of one-off reviews and the difficulty of maintaining a reliable audit trail through standard chatbot interfaces. A persistent review process integrated with engineering systems, it argues, is better suited to tracking long and interrupted security examinations.

In the two weeks after the Coldcard attack, Prem said it worked with more than 50 Bitcoin companies and processed more than 10 billion tokens of security research. That work uncovered three critical vulnerabilities, which were disclosed to maintainers and fixed.

Two Bitcoin companies, Breez and ArkLabs, tested CyberScan on production code during its beta period. Prem said ArkLabs now uses the platform on every software release.

Roy Sheinfeld, Co-founder and CEO of Breez, described how the company has used the product in recent weeks.

"The recent wave of AI-driven attacks on Bitcoin projects made one thing clear: security reviews need to move at the same speed as the threats. We've been using CyberScan for the past few weeks, and it's become a genuinely useful layer in how we review our code. Bitcoin infrastructure demands this level of rigor, and we're glad Prem is building for it," Sheinfeld said.

ArkLabs said it helped shape the product while validating it against its own codebase.

"We got early access to CyberScan, and together with the Prem team we validated it against our own code, with our feedback helping shape the product along the way. Today, every release we ship goes through CyberScan, and our code never leaves our control," said Marco Argentieri, CEO of ArkLabs.

How it works

Teams connect CyberScan to a repository, where the system maps the codebase, examines possible attack paths and produces structured findings. Prem said those findings can be formatted as GitHub code-scanning alerts and issues, avoiding the need for developers to copy conclusions from chat transcripts into other systems.

Prem is offering the product through an API and through the Model Context Protocol, or MCP, allowing engineering teams to connect it to existing development tools. It also supports model selection for each scan, with different systems available for release audits and routine scanning.

Alongside CyberScan, Prem has introduced a version of DeepSeek V4 Flash 0731 tailored for security review tasks. The model is available as an option in CyberScan at launch.

Bitcoin first

Prem is starting with Bitcoin because open-source financial infrastructure is both a visible target and a difficult test case for security tools. The company said the same broad categories of software defects seen in cryptocurrency systems can also appear in fintech and banking code.

Simone Giacomelli, Founder and CEO of Prem AI, said the product reflects a wider shift in how open-weight AI models are being used in defensive security work.

"Open-weight models have quietly become some of the best security reviewers in the world. Pair them with encrypted inference and they can do that work autonomously and continuously, on legacy code and new code alike, without a company ever giving up control of it. That combination is what CyberScan is built on. Bitcoin is the canary in the coal mine. It gets hit first, and the rest of financial services will follow," Giacomelli said.