Pi

Pi, also known as Pi Security, is an agentic product security company based in San Francisco and founded in 2025. Chief executive Guy Arazi spent years as a security researcher at Microsoft, working on Defender, Azure, and the Microsoft Security Response Center, and earlier helped establish cloud security work at Palo Alto Networks. Chief product officer Yoni Ramon led offensive security at Tesla for more than a decade across the company's vehicles and robotics programs. In June 2026 the company announced thirty five million dollars led by Brightmind Partners and Third Point Ventures, with backing from CrowdStrike chief executive George Kurtz and the Armis founders. Pi's premise is that AI has made finding vulnerabilities cheap while fixing them remains the hard part, because a fix must withstand attack, preserve functionality, and address every similar weakness.

Pi acts like a senior security engineer with deep knowledge of a company's development environment. At its center is a security brain, an institutional memory that ingests codebases, cloud infrastructure, design documents, engineering discussions in Slack and Microsoft Teams, and the history of past incidents and how they were resolved. Grounded in that context, the platform distinguishes vulnerabilities that pose real risk from the false positives that overwhelm code only scanners, then ships fixes autonomously across the software development lifecycle, from design through production. As companies increasingly build with AI coding agents, Pi gives those agents the institutional knowledge to understand how the organization builds, and breaks, its own software, so what they ship stays secure.

The company reports early traction with a demanding customer base that it describes rather than names, among them a frontier AI lab, cybersecurity firms, a major social network, and leaders in travel and insurance. For an engineering organization shipping quickly with AI in the loop that wants remediation grounded in its own history rather than a wall of generic findings, Pi is a strong fit. A team that only needs point in time scanning, or that requires an air gapped self hosted deployment or open model choice, will find Pi's context driven, cloud delivered model less aligned with those constraints, since those options are not documented.