Cyber Threat Brief — Issue #010

A criminal group has been systematically collecting the login credentials of Fortinet VPN devices used by organizations worldwide. The operation, called FortiBleed, has targeted more than 73,000 Fortinet firewalls across 194 countries and confirmed stolen credentials from over 30,000 of them, including government agencies, corporations, and critical infrastructure operators. The group tried more than a billion password combinations against Fortinet devices and more than two billion against Microsoft database servers. When they find a match, they use the stolen credentials to get inside the organization's internal network, often without triggering any alarms. The attack does not require hacking anything in the traditional sense. It uses passwords that were already stolen in previous breaches and leaked online, tries them against internet-facing Fortinet devices, and walks through the door when one works. The size of this operation means that if your organization runs a Fortinet VPN and has not changed its passwords recently, there is a meaningful chance those credentials are already in this group's possession.

Attackers are actively breaking into Fortinet's FortiSandbox product through three separate security flaws simultaneously. FortiSandbox is the component that other Fortinet security products rely on to decide whether a file or website is dangerous. When it gets compromised, the entire security stack it supports can be blinded. Two of the three flaws had patches available since April. Attacks began in early June, two months after the fixes were released. The third flaw was patched only last week, and attackers were attempting to exploit it within days using code that appears to have been written with AI assistance. Researchers noted the AI-generated exploit was poorly written and likely faulty, but attacks still found partial success against unpatched systems. That detail matters: attackers are now using AI to write attack code faster than defenders can patch, and even broken AI-generated exploits are finding targets. If you run FortiSandbox and have not applied the latest updates, you are currently exposed to active attacks.

Today, attackers compromised a developer account and used it to publish 143 malicious software packages targeting developers who build AI applications with a popular tool called Mastra. The attack was subtle. The Mastra code itself was not changed. Instead, attackers added a single line to the package instructions that told any developer who installed Mastra to also download a hidden malicious package called easy-day-js. That hidden package was designed to steal AI service credentials, cloud access keys, and other sensitive information from the developer's machine and build environment. Because Mastra is a framework for building AI agents, the credentials it has access to are particularly valuable. They include the API keys that connect to services like OpenAI, Anthropic, and cloud providers. The attack was caught within hours by multiple security firms. But any developer who ran a Mastra installation today before the alert went out may have already had credentials stolen without knowing it.