# Agentic backdoors via pretraining poisoning

- HackerByte page: <https://hackerbyte.net/en/stories/the-creation-of-security-vulnerabilities-in-large-language-models-llms-by-intent-50341334>
- Original source: <https://arxiv.org/abs/2610.37741>
- Hacker News discussion: <https://news.ycombinator.com/item?id=50341334>
- Submitted to Hacker News: 2026-11-02T18:30:58.000Z

The creation of security vulnerabilities in Large Language Models \(LLMs\) by intentionally injecting malicious inputs into their training data.

## AI-generated summary

No summary is available in this language yet.