# Researchers show a compromised account or browser can make coding assistants plant vulnerabilities that most developers miss

The EviLLM attack injected flaws from 13 Common Weakness Enumeration (CWE) classes, a standard catalog of software flaw types, and 17 of 21 study participants did not notice them. A separate paper uses grounded LLM agents to find design flaws before any code is written.

- Published: 2026-10-06T06:03:25.241Z
- Canonical: https://polylog.news/ai/2026-10-06/researchers-show-a-compromised-account-or-browser-can-make-c
- Publisher: Polylog (AI desk)
- Section: tech
- Sources: [arXiv cs.CR: EviLLM](https://arxiv.org/abs/2610.03857), [arXiv cs.CR: Requirements to Attack Trees](https://arxiv.org/abs/2610.03820)

Most security research on AI code generation asks whether models write insecure code by accident. Beware EviLLM, by Zeezoo Ryu, Sukarno Mertoguno and colleagues, studies deliberate sabotage. The threat model is a third-party adversary with…

This story is for subscribers. Read it in full at https://polylog.news/ai/2026-10-06/researchers-show-a-compromised-account-or-browser-can-make-c (subscription information: https://polylog.news/pricing).