- Enkrypt AI introduces LLM Safety Leaderboard at RSA conference to ensure secure integration of Generative AI in enterprises.
- The leaderboard offers insights into vulnerabilities and hallucination risks of LLMs, aiding informed decision-making for technology teams.
- Key features include comprehensive Vulnerability Insights and Ethical/Compliance Risk Assessment.
- Integrated within Enkrypt’s Sentry suite, the leaderboard complements existing offerings for holistic LLM management.
- Enkrypt’s preprint paper highlights increased LLM vulnerabilities from common practices, mitigated by Sentry Guardrails.
- CEO Sahil Agarwal emphasizes commitment to secure AI integration, while CTO Prashanth Harshangi underscores proactive risk management.
Main AI News:
The exponential integration of Generative AI, even within regulated environments, continues to elevate concerns regarding the security and integrity of Large Language Models (LLMs) among cybersecurity experts. Globally, policymakers and security practitioners are actively seeking innovative technologies to address the inherent risks associated with Generative AI advancements. Just recently, the US Department of Homeland Security established an advisory board to examine the impact of artificial intelligence on critical infrastructure.
In the corporate landscape, LLMs are increasingly recognized as potential drivers of backend operations, facilitating data processing and expediting decision-making at the frontline. Take, for instance, a fintech organization leveraging an LLM-powered application to assess loan applications. However, such implementations raise apprehensions about implicit biases, as LLMs often mirror societal disparities inherent in their training datasets sourced from the internet. Moreover, instances such as Google’s LLM displaying biased behaviors underscore the hazards of inadequately addressing these biases. Questions regarding the safety of Anthropic’s Claude3 Model or the readiness of Cohere’s Command R+ LLM for enterprise utilization further emphasize the critical necessity for rigorous evaluations to prevent the reinforcement of societal inequalities and potential harm.
At the highly anticipated RSA conference, Enkrypt AI, a pioneer in securing Generative AI technologies, will unveil its latest breakthrough: the LLM Safety Leaderboard. This innovative offering forms part of Enkrypt AI’s comprehensive Sentry suite, engineered to empower enterprises with enhanced security measures and confidence.
The LLM Safety Leaderboard serves as a pivotal tool furnishing invaluable insights into the vulnerabilities and hallucination risks associated with various LLMs. This resource equips technology teams with the knowledge required to make informed decisions concerning the selection of models aligning with their specific requirements. By fostering awareness and understanding of the relative strengths and weaknesses of diverse LLMs, this tool empowers AI engineers to make judicious choices based on each model’s unique attributes.
Key highlights of the LLM Safety Leaderboard encompass comprehensive Vulnerability Insights, offering meticulous assessments of potential security threats, encompassing data breaches, privacy infringements, and susceptibility to cyber intrusions. Additionally, the Ethical and Compliance Risk Assessment evaluates biases, toxicity levels, and adherence to ethical standards and regulatory mandates, ensuring alignment with enterprise values and regulatory directives.
Integrated within Enkrypt’s Sentry suite, the LLM Safety Leaderboard complements existing offerings such as Sentry Red Team, Sentry Guardrails, and Sentry Compliance. This cohesive suite embodies a holistic approach to managing and fortifying LLMs, adhering to the highest standards of privacy, security, and compliance within enterprise ecosystems.
This unveiling coincides with the release of a groundbreaking preprint paper by Enkrypt AI titled “Increased LLM Vulnerabilities from Fine-tuning and Quantization,” which sheds light on the heightened security risks resulting from common business practices like fine-tuning and quantization, notably the increased susceptibility to jailbreaking attacks. However, the implementation of external guardrails platforms like Enkrypt’s Sentry Guardrails has proven effective in mitigating such vulnerabilities. In fact, Enkrypt’s Sentry Guardrails yielded a remarkable 9x reduction in vulnerability to jailbreaking attacks in one instance.
Sahil Agarwal, CEO of Enkrypt AI, affirmed, “The launch of the LLM Safety Leaderboard underscores our commitment to facilitating the secure and responsible integration of generative AI in the enterprise landscape. This tool serves as an indispensable resource for organizations navigating the intricacies of AI adoption, instilling confidence in their security posture.”
Echoing this sentiment, Prashanth Harshangi, CTO of Enkrypt AI, remarked, “Over the past two quarters, our dedicated team has focused exclusively on enhancing generative AI safety, culminating in significant strides with our Sentry Suite. Comprising Sentry Red Team, Sentry Guardrails, and Sentry Compliance, our suite is tailored to identify potential risks and empower businesses to proactively manage and mitigate challenges, thereby facilitating informed decision-making.”
Conclusion:
Enkrypt AI’s introduction of the LLM Safety Leaderboard signifies a pivotal advancement in ensuring the secure and responsible adoption of Generative AI technologies within enterprise environments. By addressing critical concerns surrounding LLM security and offering comprehensive insights into model vulnerabilities, Enkrypt empowers organizations to make informed decisions, ultimately fostering trust and confidence in AI integration strategies. This development underscores a growing market demand for robust AI security solutions and proactive risk management frameworks, positioning Enkrypt as a leader in this evolving landscape.
