CalypsoAI has recently unveiled the CalypsoAI Security Leaderboard, an innovative index that ranks major AI models based on their security performance. This leaderboard assesses the ability of AI models to withstand advanced security attacks and provides a risk-to-performance ratio as well as a valuable cost of security metric.
The development of the CalypsoAI Security Leaderboard was fueled by the extensive testing of AI models with CalypsoAI’s new Inference Red-Team solution. This solution combines Agentic Warfare with automated attacks to stress-test the security capabilities of various AI models.
Donnchadh Casey, the CEO of CalypsoAI, emphasized the importance of understanding the risks associated with adopting AI in organizations. He highlighted the significance of the CalypsoAI Security Leaderboard as a benchmark for business and technology leaders to safely integrate AI on a large scale.
The Inference Red-Team product by CalypsoAI offers automated assessments that simulate real-world attacks to identify vulnerabilities proactively. By utilizing Agentic Warfare and a comprehensive library of signature attacks, organizations can enhance governance, ensure compliance, and maintain secure and resilient AI systems.
Amit Levinstein, VP Security Architecture & CISO at CYE, lauded CalypsoAI’s red teaming solution as a groundbreaking advancement in AI security. He emphasized the importance of having concrete evidence and confidence to deploy AI applications securely.
With a team boasting over 70 years of combined experience in security and AI, CalypsoAI recognized the need for actionable reports that identify vulnerabilities in AI systems. This enables security teams to fortify their AI systems and stay ahead of evolving threats.
James White, President and CTO of CalypsoAI, highlighted the significance of Agentic Warfare in uncovering security gaps in GenAI models. He underscored the value of the CalypsoAI Security Leaderboard in bridging the gap in publicly available information on model security.
The dynamic nature of the AI threat landscape necessitates robust testing of AI systems to uncover hidden weaknesses. CalypsoAI’s Inference Red-Team leverages AI-powered adversaries through Agentic Warfare to engage dynamically and expose vulnerabilities that static tests may overlook.
Jay Choi, CEO of Typeform, commended CalypsoAI’s Red Team as a game-changer for businesses exploring AI initiatives. He emphasized the elimination of the trade-off between security and innovation, providing a secure framework for integrating AI technology.
The introduction of the CalypsoAI Security Index (CASI) addresses the complexity of evaluating model security. By assigning a CASI score to each model, users can gauge the security level of different AI applications. This metric, updated quarterly, offers a nuanced approach to assessing security beyond traditional metrics like Attack Success Rate.
CalypsoAI continues to collaborate with model providers to enhance security and address vulnerabilities responsibly. By updating CASI scores regularly, users can make informed decisions about the security of AI models and applications.