New TokenBreak Attack Bypasses AI Moderation with Single-Character Text Changes

Cybersecurity researchers have discovered a novel attack technique called TokenBreak that can be used to bypass a large language model's LLM safety and content moderation guardrails with just a single character change. "The TokenBreak attack targets a text classification model's tokenization...

Detecting Zero-Day Web Attacks with an Ensemble of LSTM, GRU, and Stacked Autoencoders

The rapid growth in web-based services has significantly increased security risks related to user information, as web-based attacks become increasingly sophisticated and prevalent. Traditional security methods frequently struggle to detect previously unknown zero-day web attacks, putting sensitiv...