Status:

• Head of AI at Cracken; passive security skeptic seeking trust-minimized, assumption-light, non-heuristic security & privacy (but, in reality, I spend most of my time either breaking ML models or building ML models to break stuff).
• Working on:
  • LLMs & Agentic stuff Security (yes, I know..)
  • Machine Learning Privacy
  • Password Security ❤️

Preprints:

• Dario Pasquini, Michal Bazyli, Taras Fedynyshyn, Artem Sorokin
  Red-Teaming the Agentic Red-Team
• Dario Pasquini, Evgenios M. Kornaropoulos, Giuseppe Ateniese.
  Hacking Back the AI-Hacker: Prompt Injection as a Defense Against LLM-driven Cyberattacks
  Media coverage: Dark Reading, Schneier on Security

Blogposts:

• Dario Pasquini: Lying Your Way Out of the (Claude Code) Sandbox
• Dario Pasquini, Petros Efstathopoulos: Indirect Prompt Injection and filters bypass on Apple Intelligence
  Media coverage: The Register, Fox News

Works I like to talk about:

• [USENIX Sec'26] -- Dario Pasquini, Evgenios M. Kornaropoulos, Giuseppe Ateniese, Omer Akgul, Athanasios Theocharis and Petros Efstathopoulos When AIOps Become "AI Oops": Subverting LLM-driven IT Operations via Telemetry Manipulation -[Media coverage: The Register, Schneier on Security]
• [USENIX Sec'25] -- Dario Pasquini, Evgenios M. Kornaropoulos, Giuseppe Ateniese.
  LLMmap: Fingerprinting For Large Language Models
• [ACM AISec 24] -- Dario Pasquini, Martin Strohmeier, Carmela Troncoso.
  Neural Exec: Learning (and Learning from) Execution Triggers for Prompt Injection Attacks
• [IEEE S&P'24] -- Dario Pasquini, Danilo Francati, Giuseppe Ateniese, Evgenios M. Kornaropoulos.
  Breach Extraction Attacks: Exposing and Addressing the Leakage in Second Generation Compromised Credential Checking Services
  
• [IEEE S&P'24] -- Dario Pasquini, Giuseppe Ateniese, Carmela Troncoso.
  Universal Neural-Cracking-Machines: Self-Configurable Password Models from Auxiliary Data
• [IEEE S&P'23] -- Dario Pasquini, Mathilde Raynal, Carmela Troncoso.
  On the (In)security of Peer-to-Peer Decentralized Machine Learning
• [ACM CCS'22] -- Dario Pasquini, Danilo Francati, Giuseppe Ateniese.
  Eluding Secure Aggregation in Federated Learning via Model Inconsistency
• [ACM CCS'21] -- Dario Pasquini, Giuseppe Ateniese, Massimo Bernaschi.
  Unleashing the Tiger: Inference Attacks on Split Learning
• [USENIX Sec'21] -- Dario Pasquini, Marco Cianfriglia, Giuseppe Ateniese, Massimo Bernaschi.
  Reducing Bias in Modeling Real-world Password Strength via Deep Learning and Dynamic Dictionaries
• [IEEE S&P'21] -- Dario Pasquini, Ankit Gangwal, Giuseppe Ateniese, Massimo Bernaschi, Mauro Conti.
  Improving Password Guessing via Representation Learning

Invited talks (the ones I remembered to put here):

• "Red-Teaming the AI-Red-Team" [ELSA General Assembly workshop 🎥]
• "Hacking-back the AI-Hacker (Mantis)" [RSAC conference 🎥]
• "Neural Exec: Learning Execution Triggers for Prompt Injection Attacks" [Swiss AI Sec. Network, Sapienza University]
• "Privacy Preserving Collaborative Machine Learning?" [CNR Italy, Inria Rennes, Monash University, Roma3 University, JPMorgan AlgoCRYPT Center of Excellence]
• "Password Alchemy with Universal Neural-Cracking-Machines: Transmuting email addresses into Password Models" [PasswordsCon'23, Roma3 University]

LLM sec projects:

• AIOpsDoom: Prompt injection on AIOps made easy
• Project Mantis: Defending from LLM-driven cyberattacks using prompt injection
• LLMmap: like nmap but for LLMs...
• NeuralExec: Toolset to generate highly effective (optimization-based) prompt injections

Program Committees:

• IEEE S&P 2026
• ACM CCS 2023/2025/2026
• USENIX Sec 2023/2025
• IEEE SaTML 2024/2025/2026
• PETS 2025