Projects
Benchmarking Prompt Injection Attacks Against Large Language Models (LLMs).
Poster in Docs, Click Link!
Benchmarking Prompt Injection Attacks Against Large Language Models (LLMs). - by Aurelia Brzezowska
Qwen2 and Llama3.2
Benchmarking Prompt Injection Attacks Against Large Language Models (LLMs). - by Aurelia Brzezowska
Llama3.1 and Gemma2
Benchmarking Prompt Injection Attacks Against Large Language Models (LLMs). - by Aurelia Brzezowska
ASR versus MR
Benchmarking Prompt Injection Attacks Against Large Language Models (LLMs). - by Aurelia Brzezowska
Artefact Design Diagrams
I’m Aura, a national award-winning, aspiring Cyber Security graduate and chair of the Youth Advisory Panel for the UK Cyber Security Council.
My project explores a gap in AI security: whether quantisation affects a model’s vulnerability to direct prompt injection attacks. As generative AI becomes more accessible, businesses must be aware of the risks behind deployment choices.
While quantisation can reduce computational costs, does it also change how secure a model is? I investigate whether different precisions levels (e.g. FP16, INT8, INT4) influence model security, creating an open-source benchmark for comparison to reuse for future models across various precisions levels.
https://www.linkedin.com/in/aurelia-brzezowska/
Contact Aurelia