Key Takeaways
- Ambient prevents physical security incidents by proactively using AI to monitor camera feeds for large enterprises.
- CEO Shikhar Shrestha's traumatic childhood experience with an unmonitored security camera motivated Ambient's mission.
- Ambient developed Pulsar, a proprietary Vision Language Model (VLM) optimized for security camera footage.
- The future of physical security involves AI detecting threats and automatically responding to incidents.
- Selling AI software into the physical security market requires educating clients accustomed to hardware-centric solutions.
Deep Dive
- Ambient's mission is to prevent physical security incidents by using AI to monitor camera feeds and alarm events in real-time for large enterprises.
- CEO Shikhar Shrestha's personal motivation stems from an armed robbery at age 12 where a security camera was present but unmonitored.
- The AI identifies precursor activities or suspicious behaviors that precede incidents, enabling proactive intervention rather than reactive responses.
- AI's role in security encompasses prevention, response, and forensics, detecting suspicious activity before incidents and assisting post-incident investigations.
- Ambient's core vision-language model technology, built since 2016, has been significantly enhanced by recent Generative AI and LLM breakthroughs.
- Advanced Vision Language Models (VLMs) can differentiate genuine threats from normal activities, such as distinguishing a medical fall from car repair or vandalism from drawing on a whiteboard.
- Ambient developed its own VLM, Pulsar, due to the high cost of running large public models on continuous video feeds (estimated $5,000-$10,000 per camera monthly) and their ineffectiveness with security camera imagery.
- Pulsar is 50 times more compute-efficient and outperforms public models for threat detection in security footage, trained on proprietary data from tens of thousands of deployed cameras.
- Ambient addresses privacy by avoiding facial recognition, focusing on detecting suspicious behaviors and precursor actions, and prompting the AI with specific threat-related questions.
- The future of AI in physical security is seen as progressing to real-time assessment of detections within site-specific contexts and culminating in automated responses.
- Automated actions could include locking down facilities, notifying law enforcement, and performing remediation, with customer desire driven by the critical importance of seconds during crises.
- Ambient already provides capabilities like license plate detection for exterior cameras and can connect various data points for end-to-end forensic investigations.
- Ambient's AI detected a spontaneous outdoor battery fire at an aerospace manufacturing facility, alerting the customer within a minute and preventing a major plant shutdown.
- The system also detected a fence line breach and subsequent break-ins at an e-commerce company's site, leading to the apprehension of the perpetrator within minutes of the alert.
- Selling software in the physical security industry involves educating clients on 'agentic physical security' and differentiating it from traditional manual operations and hardware-focused solutions.
- Initial fundraising for Ambient was challenging due to past failures in physical security tech, but investor interest has since grown due to the market's size and manual operations.
- During COVID-19, Ambient repositioned its product for active verticals like museums and data centers, observing a post-pandemic shift towards remote monitoring and cloud-based security solutions.