PromptHalo

AI systems don’t fail at the model layer—they fail in what they are allowed to do in real environments. 
As AI systems become agentic, they move from generating outputs to taking actions across systems, data, and workflows. This shifts risk from model quality to decision execution. Existing security controls were not designed for this layer.

PromptHalo addresses this by governing AI behavior at runtime. We sit between AI systems and execution, enforcing what agents are allowed to do and preventing unintended or out-of-scope actions. This is emerging as a control plane problem, not just a detection problem.

 While many of the AI runtime security solutions focus primarily on the prompts, and the models, we focus on controlling actions and outcomes of the agentic applications for reliable innovation - especially in financial services industry. 

We offer SaaS service for most customers and a hosted model for enterprise customers looking for private hosting. 

PromptHalo was founded by Madhuri Chandoor in 2025. Her background in enterprise and financial systems, where control, auditability, and risk management are non-negotiable, informs how we approach AI behavior governance. She founded PromptHalo, to solve a problem she has seen repeatedly in financial institutions - as systems become more intelligent and interconnected, the attack surface grows faster than security teams can understand or control it. 

The team has  experienced engineers including a security engineer, and two of Madhuri's ex-AWS colleagues. PromptHalo team has experienced advisors with collective experience of about 60+ years building enterprise scale applications at PayPal, Google and AWS,  guiding and vetting the approach along the way.