Generative AI and Cybersecurity: SydeLabs' Innovative Solutions
Generative AI is transforming modern businesses, but it also introduces new attack vectors that can jeopardize both organizations and their reputations. California-based startup SydeLabs is addressing this challenge with a real-time, intent-based firewall. Recently, the company secured $2.5 million in seed funding from RTP Global, Picus Capital, and various angel investors.
While multiple players exist in the AI security sector, SydeLabs differentiates itself with a comprehensive suite of solutions designed to help developers avoid vulnerabilities in large language models (LLMs), including lesser-known risks throughout the project lifecycle—from development to deployment.
Innovative Product Offerings
SydeLabs has developed three key AI-focused products: SydeBox, SydeGuard, and SydeComply. SydeBox, currently in beta, serves as a self-service red-teaming solution that allows teams to evaluate their AI applications and models for potential vulnerabilities. SydeGuard, set to launch soon, provides real-time intent-based protection, while SydeComply addresses compliance issues across different global regions.
“During SydeBox's beta phase, we uncovered various vulnerabilities, including training data leaks, prompt injections, and safety alignment bypasses,” said Ankita Kumari, co-founder of SydeLabs. “SydeGuard can detect and prevent prompt injections, denial of wallet attacks, and data leak attempts.”
How SydeLabs Works
SydeBox utilizes an AI agent that initiates test attacks based on both in-house research and publicly available data, while an LLM assesses the attack’s success as it interacts with the target system. Conversely, SydeGuard employs proprietary models that analyze user prompts for potential threats, assigning risk scores to prompts, user sessions, and users.
Instead of blocking users immediately, SydeGuard informs enterprise security teams, offering options to block, monitor, or redirect the threat to a honeypot, allowing teams to decide how to address potential attacks strategically.
Robust Vulnerability Detection
SydeLabs has flagged over 10,000 vulnerabilities in more than 50 applications and models since launching the beta of SydeBox on March 1, 2024. As the company continues to grow, it aims to finalize its offerings of SydeGuard and SydeComply. Long-term, SydeLabs plans to provide the red teaming solution for free to enterprises focused on vulnerability detection while monetizing SydeGuard through a consumption-based model.
Kumari emphasized the importance of a cybersecurity product’s accuracy and its capacity to stay ahead of cyber threats. “We prioritize constant upgrades to our model, focusing on new attack vectors and methodologies.”
Industry Context and Competitive Edge
SydeLabs' funding comes at a time when the security of generative AI applications is under intense scrutiny, highlighted by incidents of prompt injection attacks resulting in deepfakes of celebrities. In response, Microsoft has launched new Azure AI tools aimed at detecting hallucinations and preventing attacks. Competitors like Lakera and Prompt Security operate in the red teaming and real-time prevention space, but Kumari asserts that SydeLabs’ products demonstrate superior accuracy and performance compared to these tools.
SydeLabs continues to innovate within the cybersecurity landscape, ensuring generative AI systems remain secure against evolving threats.
Ranking
