Pillar Security raises $9M in Seed funding to tackle AI-specific cyber threats
With attacks now targeting LLMs and training data, Pillar is building the defense layer for agentic software.
As companies race to integrate artificial intelligence into everything from customer service to code generation, the speed of innovation is outpacing traditional security controls. Now, a startup is betting that security in the age of agentic, self-executing software will require a ground-up rethink—not just an update.
Full list of Israeli high-tech funding rounds in 2025
Pillar Security, an Israeli startup focused on end-to-end AI application protection, has emerged from stealth with $9 million in Seed funding to tackle what it sees as a widening security gap in the enterprise AI stack. The round was led by Shield Capital and joined by Golden Ventures, Ground Up Ventures, and a group of strategic angel investors.
Founded in October 2023 by Dor Sarig and Ziv Karliner, Pillar aims to become the foundational security layer for AI-native software—particularly as companies deploy agentic systems capable of autonomous decision-making and interaction.
“AI is fundamentally changing the way we build software — it doesn't just add another step to traditional processes; it introduces an entirely new lifecycle.” said Dor Sarig, CEO & Co-Founder at Pillar Security. “In the intelligence age, data is executable and software has agency. Pillar’s technology, backed by real-world AI threat intelligence, is built with this understanding, delivering a new class of protection designed explicitly for AI-related security risks. We are redefining application security to match the agentic and autonomous software of the Intelligence Age.”
Related articles:
According to a recent Deloitte survey of 1,200 cybersecurity leaders, two in five organizations have already experienced AI-related security or privacy incidents, including model manipulation, data leakage, and IP theft. In 25% of those cases, the threat was malicious.
Traditional tools are ill-equipped to identify and remediate attacks like jailbreaks on LLMs (large language models), prompt injections, or data poisoning of training sets—attacks that don’t always follow recognizable patterns.
Pillar's approach is to treat AI systems as dynamic, risk-sensitive assets, mapping everything from the data and prompts that train them to the run-time infrastructure they depend on. From there, the platform performs red-team simulations, builds adaptive guardrails, and continuously assesses operational risk based on live threat telemetry. The startup claims it can detect and prevent AI-targeted attacks with far more granularity than incumbent security software.
