World Congress 2026 Europe Jul 10, 2026 Session details

The Sound of Your Secrets: Teaching Your Model to Spy, So You Can Learn to Defend

David vonThenen

Did you know your microphone can read your typing? Discover how deep learning translates acoustic keystrokes into text, and learn to defend against these sophisticated side-channel attacks.

Pause
Mute Enter Fullscreen
#1 about 2 min

Understanding the concept of acoustic keystroke loggers

Translating individual typing sounds into recognized characters using specialized machine learning.

#2 about 2 min

Moral duality of artificial intelligence and machine learning

Comparing the societal benefits of algorithm research against the manipulative capabilities of advanced optimization.

#3 about 3 min

Translating audio signals into spectrographic images for classification

Converting raw sample recordings into distinct visual heatmaps that highlight unique frequency intensities.

#4 about 3 min

Processing audio datasets into training and testing splits

Preparing noiseless and multiple device audio collections to construct validation checkpoints.

#5 about 2 min

Demonstrating the audio to spectrographic image conversion pipeline

Executing python commands to generate visual rgb images from captured linear wave files.

#6 about 2 min

Training the classification model and running single keyboard inference

Loading a pre-trained safetensor pipeline to accurately pair acoustic fingerprints with their specific input characters.

#7 about 4 min

Running acoustic inference across a multiple keyboard dataset

Evaluating detection logic against independent laptop hardware models to measure confidence distribution falloff.

#8 about 7 min

Reconstructing full sentences with acoustic combinations and spellcheck

Using the distinctive resonance of the spacebar to isolate probability chains against dictionary APIs.

#9 about 2 min

Predicting missing words using specialized language models

Applying small language context engines to automatically hypothesize dropped characters inside imperfect transcriptions.

#10 about 3 min

Analyzing model accuracy limits and manufacturing hardware similarities

Understanding why high precision requires keyboards that physically align with factory production lots.

#11 about 2 min

Exploring practical applications of acoustic side channel attacks

Combining hidden microphone bugs with skimmer overlays to steal unencrypted pin code numbers.

#12 about 2 min

Implementing defensive countermeasures against acoustic side channel vulnerabilities

Preventing effective snooping by introducing intentional multi-channel noise streams and increasing overall password entropy.

#13 about 3 min

Providing educational resources for acoustic attack defense code

Reviewing open repository references so engineers can independently replicate and evaluate hardware vulnerabilities.

Matching moments

5:11 min

Exploiting systems through audio manipulation and prompt injections

Chris Heilmann +2 · LIVE

3:16 min

Unique privacy vulnerabilities in language model architectures

Aditi Godbole · LIVE

2:12 min

Open-source voice cloning and deepfake vulnerabilities

Chris Heilmann +3 · LIVE

1:15 min

Retaining the defender advantage in the cybersecurity race

Michele Zuccala Michele Zuccala +4 · WWC Europe 2026

6:36 min

Defensive strategies against AI-driven social engineering

Wolfgang Ettlinger +1 · WWC 2023

58 sec

Manipulating AI models for privacy and anonymity

David vonThenen David vonThenen · WWC 2025

Upcoming sessions on this topic

Open session

World Congress 2026 North America

The Things Your AI Isn't Telling You

Desmond Lamptey

Lead Software Engineer @ Capital One

Desmond Lamptey
Open session

World Congress 2026 North America

SecurePrompt: Building a Pre-Flight Security Layer for Agentic AI

Ravi Sastry Kadali

AI/ML Engineer at General Motors

Ravi Sastry Kadali
Open session

World Congress 2026 North America

From Software Agents to Physical Devices: Inside the Agentic Hardware Stack

Michael Yuan, Vivian Hu

Michael Yuan
Vivian Hu
Open session

World Congress 2026 North America

Red Teaming Your LLM App -- A Hands-On Threat Model You Can Reuse

Saloni Garg

Senior ML Engineer at Adobe

Saloni Garg
Open session

World Congress 2026 North America

Secure-by-Inclusion: Preventing Accessibility Barriers from Becoming Security Vulnerabilities

Radostina (Ina) Tsvetkova

Norwegian Directorate of Labour and Welfare (NAV), Senior Advisor in Digital Accessibility and Inclusive Design

Radostina (Ina) Tsvetkova
Open session

World Congress 2026 North America

Context Engineering Kung Fu

Carl Lapierre

Tech Lead and AI Engineer at Osedea

Carl Lapierre