Cryptography — The Video Signal

Aug 19, 2026 · 47:47

How Claude's Text Watermarking Works

This video provides a deep technical explanation of how Anthropic's text watermarking for Claude models operates. The technique modifies the standard next-token sampling process by introducing determinism using a secret 'watermarking key.' Instead of relying purely on random probability distribution (e.g., via softmax and random choice), the model uses specialized functions and a tournament sampling method to ensure that certain token choices are predictable, making the generated text traceable. Detection is achieved by applying these same watermarking functions across the entire text and calculating an average score against a defined threshold.

Key takeaways

Watermarking Location 0:24
The watermark is applied at the *sampling* stage of the LLM generation process, not within the core model weights. This means existing LLMs can implement this feature without requiring full retraining or modification of the underlying model architecture. (24:45)
Deterministic Sampling 0:16
The watermarking mechanism converts standard random sampling into a deterministic process by using a secret 'watermarking key' and previous tokens to derive a fixed random seed, ensuring that the same input always yields the same sequence of watermarked tokens. (16:19)
Tournament Sampling 0:36
To make detection efficient, the process uses 'tournament sampling.' Instead of simple random choice, plausible next tokens are paired up and compared using multiple specialized watermarking functions ($G_1, G_2, ext{etc.}$), which determines the final selected token. (36:36)
Detection Method 0:47
Watermark detection is performed by running the text through the same set of watermarking functions and calculating an average score across all positions. If this average score exceeds a specific threshold, the text is flagged as watermarked. (47:13)

Watch on YouTube Full article

Aug 11, 2026 · 51:32

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

This talk details acoustic keystroke logging—a method of intercepting typed information purely from sound rather than physical interception. The speaker outlines how deep learning models can be trained using spectrographic images derived from recorded key presses to classify specific letters (e.g., 'S'). While demonstrating the high accuracy of single-keyboard attacks (100%), the presentation shows that multi-keyboard logging is challenging but feasible, especially when combined with context prediction and spell-checking algorithms. The session concludes by emphasizing defensive measures, including two-factor authentication using physical keys and implementing strong, unique, offline password policies.

Key takeaways

Acoustic Keystroke Logging Mechanism 16:22
The attack relies on machine learning audio classification. Audio files (linear 16 format) are converted into spectrographic images (frequency over time, visualized as a heatmap), which serve as the input for training models like Convolutional Neural Networks (CNNs).
Multi-Keyboard Attack Complexity 28:12
While single-keyboard classification can achieve 100% accuracy, using multiple keyboards significantly lowers confidence scores. The problem is made solvable by decomposing the text based on space delimiters and employing spell-checking/context prediction (e.g., predicting 'hello people' from partial sound inputs).
Defensive Strategies 40:50
Defense requires layered security: use physical two-factor authentication keys (like YubiKey) instead of SMS; utilize unique, complex passwords that are not known to the user; and be aware of potential signal interference or noise.

Watch on YouTube Full article