DeepKeep builds a model-agnostic, AI-native security platform that safeguards ML models from development through deployment. The platform is organized around four pillars: Continuous Risk Assessment, Automated Prevention, Real-time Detection, and Alert and Mitigation, offering penetration testing, pre-deployment hardening, online/offline attack detection, and mitigation including an AI firewall and response options. DeepKeep began by securing computer vision models and has expanded to support a variety of ML types, including large language models (LLMs) and multimodal systems. The company reports successful trials, including preventing data leakage and stopping toxic LLM responses at a large financial institution, and has conducted evaluations of models such as Meta’s LlamaV2 7B to surface vulnerabilities like prompt injection, jailbreaks, DoS, data leakage, hallucinations, and bias. DeepKeep raised $10M in seed funding in a round led by Canadian-Israeli VC Awz Ventures. The platform is deployed by leading global enterprises in the finance, security, and AI computing sectors, and the company plans to expand multilingual NLP support with an initial focus on Japanese.

Total raised

$10M

Funding rounds

1

Latest round

Seed

Latest activity

Jul 2024