Frontier Security

Frontier Security runs controlled simulations and security assessments on the most advanced AI models before they ship, hunting for ways they could be misused for malicious cyber operations. The San Francisco firm sells this red-teaming and evaluation work as a consulting service to the labs and organisations building and deploying frontier AI.

As AI systems edge closer to autonomous agents, the question of whether they can be turned into cyber weapons — or slip their controlled testing environments — has moved from theory to something labs and governments now actively assess. Recent analysis by the company on models escaping test environments points to real, present weaknesses in cyber safeguards.

The firm positions itself at the intersection of AI safety and offensive security research, working directly with leading labs to stress-test next-generation systems. Founded in 2026 and led by CEO Yaron Singer alongside co-founder Paul Kassianik, the company is still pre-funding.

Co-founded by Yaron Singer and Paul Kassianik, the team says its analyses have already been cited in technical reports from major AI labs and used by governmental bodies to assess cyber risks — early signals of credibility in a field with few genuine experts.