Locking it down: A new technique to prevent LLM jailbreaks
A new technique called LLM salting has been introduced to counteract jailbreak attacks on LLMs such as LLaMA-2-7B and Vicuna-7B. These attacks exploit the reuse of precomputed adversarial prompts across similar model deployments.
New technique detects tampering or forgery of a PDF document
With the PDF format being used as a formal means of communication in multiple industries, it has become a good target for criminals who wish to affect contracts or aid in misinformation.
Researchers claim breakthrough in fight against AI’s frustrating security hole
Google DeepMind has unveiled CaMeL (CApabilities for MachinE Learning), a new approach to stopping prompt-injection attacks that abandons the failed strategy of having AI models police themselves.
Google Researchers Claim First Vulnerability Found Using AI
Researchers from Google Project Zero and Google DeepMind discovered a vulnerability using a large language model (LLM). The vulnerability was found in SQLite, an open-source database engine, and reported to developers before its official release.
Chinese Researchers Break RSA Encryption Using Quantum Computing
Chinese researchers have claimed to successfully break RSA encryption using D-Wave's quantum computing technology, sparking concerns about the security of traditional cryptographic methods.
Innovative Approach Promises Faster Bug Fixes
Birgit Hofer and Thomas Hirsch from TU Graz have developed a new approach to speed up software bug fixes. By identifying bottlenecks in fault localization, they created a scalable solution using NLP and metrics to analyze code for faults.
Meta's AI Safety System Manipulated by Space Bar Characters to Enable Prompt Injection
A bug hunter discovered a bypass in Meta's Prompt-Guard-86M model by inserting character-wise spaces between English alphabet characters, rendering the classifier ineffective in detecting harmful content.
Firmware Update Hides Bluetooth Fingerprints
A team of researchers from the University of California San Diego has developed a firmware update to hide a smartphone's unique Bluetooth fingerprint, which can be used to track the user.
Google Introduces Project Naptime for AI-Powered Vulnerability Research
Google has developed a framework known as Project Naptime, which utilizes a large language model (LLM) for vulnerability research. The framework allows an AI agent to simulate the actions and workflow of a human security researcher.
Google's Zero-Day Hunters Test AI for Security Research
Google's Project Zero team of zero-day hunters believes that artificial intelligence (AI) can enhance automated threat identification and analysis by detecting vulnerabilities that are often missed by current tools.
Defend Against Threats with Cyber Fusion
Cyware is the leading provider of cyber fusion solutions that power threat intelligence sharing , end-to-end automation and 360-degree threat response.
Learn More
Trending Tags
