Source-reported · Independent verification status is not recorded for this article; claims remain attributed to their sources.
CISA warns of active attacks on four widely used software flaws, including AI infrastructure
U.S. cybersecurity officials say hackers are actively exploiting critical vulnerabilities in Synology DSM, LiteLLM, Apache Axis, and REDCap to deploy ransomware, mine cryptocurrency, and hijack AI infrastructure. The alert puts hospitals, research institutions, cloud providers, and smaller IT shops on notice that their everyday tools are being weaponized.
Hackers are actively targeting four software vulnerabilities with a mix of ransomware, crypto‑mining, and attacks on artificial‑intelligence infrastructure, prompting a fresh warning from U.S. cyber authorities that the holes need to be closed fast.
According to a new alert, attackers are exploiting flaws in Synology’s DiskStation Manager (DSM), the LiteLLM middleware used to connect applications to large language models, the Apache Axis web services framework, and the REDCap research data platform. Three of the bugs carry a critical severity score of 9.8 on the industry‑standard CVSS scale, indicating they are easy to exploit and can give attackers broad control.
Synology DSM runs on network‑attached storage devices widely used by small and mid‑sized businesses, schools, and some home offices for file sharing and backups. A critical flaw there can turn a company’s backup appliance into an entry point for data theft or encryption. REDCap, short for Research Electronic Data Capture, is deployed across universities and hospitals to manage sensitive clinical and survey data; a compromise could expose patient information or derail medical studies.
LiteLLM and Apache Axis speak to a newer frontier. LiteLLM helps developers route requests from their applications to underlying AI models, often running in the cloud. Apache Axis is used to build and expose web services. Exploiting weaknesses in these layers lets attackers hijack compute resources for crypto‑mining, pivot deeper into networks, or manipulate data flowing in and out of AI systems.
For hospitals, universities, and mid‑tier service providers, the stakes are practical. A ransomware hit on a REDCap instance can freeze access to years of research. A compromised Synology box can lock up backups right when an organization needs them most. And if attackers use LiteLLM flaws to seize control of AI‑related infrastructure, they can run up cloud bills, steal intellectual property embedded in models or training data, or poison the systems that automate everything from customer service to code generation.
Operationally, the pattern fits a familiar playbook. Attackers scan the internet for exposed, unpatched instances of popular software; use pre‑built scripts to exploit known vulnerabilities; then drop payloads that either encrypt data for ransom, quietly mine cryptocurrencies, or establish backdoors for later use. The fact that three of the four flaws now under active attack rate 9.8 on the CVSS scale suggests they typically require little or no authentication and can be triggered remotely.
The alert also underlines how quickly AI tooling has become part of the critical‑infrastructure conversation. LiteLLM is not a household name, but it sits in front of systems that expensive engineering teams are rapidly wiring into core business processes. If those connectors are compromised, companies may not only lose data but also unknowingly rely on outputs from models that have been tampered with.
For national security agencies and regulators, these incidents reinforce the argument that software supply chains and open‑source components need the same scrutiny as traditional industrial control systems. A single unpatched library or middleware service can provide a jumping‑off point into government networks, defense contractors, or utilities that have layered AI into their operations.
The most memorable point here is blunt: AI doesn’t just create new threats; it also inherits every old software bug in the stack beneath it.
In the near term, key indicators will be whether major cloud providers and software vendors push emergency patches and configuration guides to their customers; whether ransomware gangs begin naming victims tied to these specific platforms; and whether regulators move to fold AI‑adjacent middleware into mandatory disclosure and baseline security regimes, rather than leaving them as a blind spot between application and infrastructure.