Depthfirst raises $80M to expand AI-native security platform and train domain-specific models #AI

Artificial intelligence-native security platform startup depthfirst Inc. announced today that it has raised $80 million in new funding to train additional security models across new domains, expand its AI research team and scale up enterprise adoption.

Founded in 2024, depthfirst is taking on the issue of a rapidly changing threat landscape where software is developed faster than traditional security tools can keep up and where attackers are increasingly using AI to find and exploit weaknesses. The company aims to secure the world’s software by building tools that understand and protect code at the speed and scale of both development and attack.

Depthfirst’s General Security Intelligence platform deploys custom AI agents to analyze and interpret a company’s codebases, infrastructure and workflows. The platform also uses deep context and machine learning to detect subtle, complex vulnerabilities that traditional tools often miss.

The Series B funding round was led by Meritech Capital Partners LP, with Forerunner Ventures, The House Fund, Accel Partners LP, Box Group, Liquid 2 Ventures, Alt Capital and Mantis VC also participating. The new funding comes just under three months since depthfirst raised $40 million in Series A funding.

“Recent public-market reactions suggest investors are starting to recognize that AI will disrupt the legacy security stack,” said depthfirst co-founder and Chief Executive Qasim Mithani. “But to win in security, companies will need to deploy security-specific models in products optimized for real security workflows.”

Alongside the funding announcement, depthfirst also today announced the introduction of its first in-house security model, dfs-mini1.

The model is initially focused on securing cryptocurrency smart contracts as part of depthfirst’s broader effort to build specialized intelligence into the security platform it already delivers to customers. Dfs-mini1 was built on an open-source model, post-trained through reinforcement learning in security-specific environments and evaluated on OpenAI EVMBench, a benchmark for smart contract vulnerabilities.

In initial testing, depthfirst says that dfs-mini1 outperformed frontier models while running at 10 to 30 times lower cost. Internal evaluations also suggest that dfs-mini1 can generalize beyond smart contracts and perform better on other security tasks, evidence that its training approach transfers across security domains.

“When you own the training process, you can optimize for what actually matters in your domain,” said Chief Technology Officer Andrea Michi. “In our case, that means vulnerability detection and verification. The result is a model that can be cheaper to run, better at the task and more responsive to continued investment than a general-purpose system.”

Image: depthfirst