Key points
- Cognition chief security officer Alex Stamos says AI-driven attacks have pushed security into a 'dark age' comparable to the 1990s, per a Verge Decoder interview.
- He says a Chinese open-source tool called ARTEX, paired with open-weight models, began with three South Korean banks and hit about 200 more companies this week.
- He describes chained breaches: stolen Fireworks API keys allowed access to Cursor, where a bad plugin reached 5,000 more companies.
- He argues the 2000s-2010s security industry rested on skill and staffing assumptions that AI is eroding.
Why it matters for marketers: Marketing stacks run on AI vendors, plugins and API keys, so a supplier breach can reach your data and accounts. Vendor security reviews now matter more.
In an interview on The Verge's Decoder podcast, Alex Stamos, chief security officer at Cognition, described a month of continuous breaches across the tech industry. He pointed to hacks at AI infrastructure companies, the Binance theft, and an incident in which stolen API keys from Fireworks, an AI services provider, were used to get into Cursor. From there, he said, a malicious plugin was pushed out and reached 5,000 more companies.
Stamos argues the cause is a jump in attacker capability. Open-weight models, he says, can now automate most of the steps of an attack, from finding bugs to writing exploits in real time. He cited ARTEX, a Chinese open-source tool that he says began with three South Korean banks and has hit about 200 more companies this week. He said the Binance breach was mostly human social engineering by North Korean actors, with perhaps some AI involvement.
For marketers and content teams, the lesson is about dependencies. Teams now lean on AI platforms, coding assistants, plugins and API keys connected to CMS, analytics and ad accounts. Stamos's Cursor example shows how one vendor's compromise can spread to customers who did nothing wrong. Smaller firms, which he says often have only a handful of IT staff, are especially exposed.
Stamos is not a doomer in the long run, but he expects a difficult adaptation period. Practical steps follow from his account: inventory the AI tools and plugins your team uses, rotate and limit API keys, and ask vendors how they would notify you after a breach. These are Stamos's views and claims from an interview, not independently verified findings.
Sources
This brief is our own summary. Read the full reporting at the source.