AI Lets Attackers Automate the Cyber Kill Chain — SkimNews
Get the Tech newsletter
Daily tech — startups, AI labs, chips, the launches that shape the next decade. Free.
- Alex Stamos describes the multibillion-dollar cybersecurity industry as overdue for AI-amplified hacks and breaches, comparing the disruption to the 1990s internet frontier and warning that the current wave is only beginning.
- ARTEX, a Chinese open-source tool, can be paired with an “obliterated GLM-5.3” model to break into organizations; the attacks began with three South Korean banks and spread to about 200 more companies that week.
- Fireworks lost API keys in a breach, which attackers used to compromise Cursor and push a malicious plugin to 5,000 companies, creating 5,000 potential breaches.
- AI systems let adversaries find software flaws and write exploits in real time, inverting the old kill chain in which researchers built reusable exploits before deploying them against one or many targets.
- Russia and North Korea are among the state actors using AI to detect vulnerabilities and break into systems, but Stamos says private actors and a range of other groups are also responsible.
- GLM-5.3 can be run in an open-weight setup on a Mac Studio, so attackers need not rely on frontier models carrying safeguards or law-enforcement logs; Stamos says this also weakens the old professional etiquette around hacking.
- Binance hack was cited by Stamos as the largest heist in human history, alongside a succession of breaches of AI-related “neoclouds” and major PHP and Node bugs.
Why it matters: For cybersecurity vendors and large enterprises, AI compresses reconnaissance, exploit writing and deployment into real time, while open-weight models sidestep frontier safeguards and logs. A chain reaching 5,000 potential breaches turns a single Fireworks compromise into a systemic vendor-risk problem for defenders.
Ask SkimNews
