Great to see this story about Xutan Peng's work on language preservation

Great to see this story about our work on security vulnerabilites of LLM-based text-to-sql, work with Xutan Peng and others

⚡️Our academics have found security vulnerabilities in six commercial AI tools that can lead to poisoning of #ChatGPT (and others’) training data. But how does poisoning an AI tool even work? 🧵 1/5

We've updated the preprint - with #vulnerability demonstrations on *four* new commercial applications (namely #ChatGPT , #Text2Sql .ai, AiHelperBot, ToolSKE) in addition to the two targets in our previous version (Baidu UNIT & Ai2Sql)! #NLP roc #NLP #cybersecurity

Academics reveal new Text-to-SQL model attacks that could potentially let attackers break into sensitive databases or launch DoS attacks.

Read deatils: thehackernews.com/2023/01/new-st…

#datasecurity #infosec #cybersecurity #privacy #technology #dataprotection

Patch Tuesday; Cacti Vuln Details; Text-to-SQL Vulnerabilities
i5c.us/p8320

Text-to-SQL Vulnerabilities Allow Data Theft and DoS Attacks

rli.to/w1M8I

#infosec #cybersecurity #machinelearning #datascience #python #deeplearning #career #tech #computerscience #internetofthings #devopsengineer #career woman #datascientist

Исследователи рассказали о новых кибератаках, в которых злоумышленники используют модель Text-to-SQL для создания вредоносного кода, позволяющего собирать конфиденциальную информацию и проводить DoS-атаки. anti-malware.ru/news/2023-01-1…

Учёные провели первую в мире кибератаку с использованием искусственного интеллекта securitylab.ru/news/535550.php

Landing my research to positively impact the real world!

做了一些微小的工作

PhD Viva ✅

So Long, and Thanks for All the Fish ♥

🤯The self-compiling AI is emerging!

Pay attention to the commercial usage of code generation, and semantic parsing in dialogue systems! Our recent work demonstrates a successful attack to such commercial softwares.