Abliteration AI released a fully de-aligned model that can scan and exploit vulnerabilities in a home network, showing how AI can bypass standard cybersecurity measures. It is the company's first major cybersecurity tool since launching its platform in 2023.
The model, based on Z.ai's GLM 5.3, identified 12 hardware systems on the same network and catalogued several vulnerabilities, including misconfigured printers and IoT devices with outdated firmware. That compares with the company's previous models, which were limited to basic security checks.
The model is built on Z.ai's agentic coding framework and targets home and small business networks. Availability begins with a limited release for cybersecurity professionals and developers.
"You have all these critical infrastructure companies, from airlines to banks, that are rolling out agents like crazy," said Devon, Abliteration AI's CEO. He added that making de-aligned models widely available is a smart defense strategy.
The announcement follows growing concerns about AI's cybersecurity risks, as companies like Anthropic and OpenAI have also released models with reduced guardrails.
Abliteration AI did not say how it will regulate access to its models, and raised questions about the ethical implications of giving users full hacking capabilities. The company plans to release more tools for network security in the coming months.
Source: wired
