Research Intern in Agentic AI Security Frameworks

Zürich - Switzerland

Monthly Salary: Not Disclosed

Posted on: 7 hours ago

Vacancies: 1 Vacancy

Job Summary

About Huawei
Huawei is a global leader in information and communications technology (ICT) renowned for its pioneering work in AI hardware large-scale computing infrastructure and integrated software-hardware solutions. With a workforce of over 194000 employees across more than 170 countries Huawei operates the worlds largest R&D organization including advanced research centers dedicated to next-generation AI and processor technologies.
At Huawei innovation isnt just a buzzwordits built into the DNA of the company. Its full-stack AI ecosystem spans from the hardware accelerators architecture to the firmware system integration workload scheduling all the way to algorithm optimizations. Meanwhile applications cover a wide range of scenarios from wearables all the way to entire clusters & data centers.

About the lab
With more than 20 sites across Europe and over 1500 researchers Huaweis European Research Institute (ERI) oversees fundamental and applied technology research academic research cooperation projects and strategic technical planning across our network of European R&D facilities.
This specific role is based at Huaweis Zurich Research Center launched 6 years ago and is already home to more than 160 experts. You would be joining the AI Computing Group within the Computing Systems Lab a dynamic team of 20 researchers focused on advancing AI solutions across hardware systems software and algorithms. Join us at the forefront of AI computing systems innovation!

Problem Statement
AI models that can meticulously reason prepare multi-stage plans and execute them to achieve complex tasks are fundamental for building future AGI systems. It is crucial that the LLM models not only generate very high-quality planning and responses but also that they can prepare code snippets to execute test self-reflect or even reach out to the outside world to not only ensure the correctness of the response but also ensure quasi-human level intelligence to guide us in achieving complex tasks. However an AI model that can execute code or communicate with the outside world (e.g. the internet) poses a monumental security threat. Recent research shows that there exist numerous ways to divert a safe response of an AI agent by either influencing its alignment generate malware or use unsafe APIs or use excessive permissions to leak sensitive user information to the outside world (e.g. directly post user query to a social media website) polluting long term model context generate malicious code and so on.
Therefore the security and safety of the current LLM-based agentic systems are questionable and is an open research question. Together we will investigate the broader security aspect of the agentic systems and design and verify the fundamental building blocks necessary for a trustworthy AI agent system.

Responsibilities

We are designing next-generation trustworthy reasoning agentic systems and investigating potential attack surfaces to mitigate them and specifically building the algorithms tools and systems for efficient and highly secure agents that can solve complex tasks by information retrieval code generation tool calling communicating with other agentic systems and so on.

As a research intern your responsibility is to contribute to these research endeavors by prototyping new security primitives for AI agents. This includes but is not limited to new OS hypervisor driver modules to run isolated execution environments identify and attack simulations. You will also be involved in rigorous security analysis and formal/semi-formal verification of the secure agentic summary you will contribute to fundamental AI security research prototyping publishing research papers at top-tier AI and security venues and drafting patents.

Requirements

You are an enrolled masters or PhD in computer science specifically in either AI/systems/ security at a reputable university or you have finished a masters or PhD very recently.

Very good understanding of AI frameworks (PyTorch/HuggingFace).

Some familiarity with attacks on AI models: membership inference prompt injection etc. is a plus.

Proficient in programming such as Python and/or C.

General Linux power user skills are an asset.

Ability to work independently on nontrivial analysis and development tasks

Strong communication skills ability to perform and present a detailed analysis of experimental results

Strong motivation to join a cutting-edge industrial research environment

By applying to this position you agree with our PRIVACY STATEMENT. You can read in full our privacy policy at