Casco

What do they actually do

Casco builds an autonomous security tester for AI agents and apps. It automatically probes LLM-powered systems with realistic attack scenarios (like prompt injection, data exfiltration, tool/plugin abuse, and system hijacks), then reports concrete, reproducible failures and how to fix them. The output looks closer to a pentest report than a static scan, with steps to reproduce and remediation guidance so engineering teams can quickly verify and patch issues Casco site.

Today, Casco is positioned as a faster, continuous alternative to manual red-teaming and pentesting for AI systems. The founders also point to early findings of novel hijack vectors in pilots, underscoring the need for targeted, AI-specific testing rather than generic scanners founder note; YC highlights that most enterprises report at least one AI-related incident, pushing demand for purpose-built testing and evidence YC profile.

Who are their target customer(s)

• Product and engineering teams building AI agents and apps: Agents can behave unexpectedly or be hijacked; teams need a way to reliably find, reproduce, and fix exploitable failures before shipping so regressions don’t reach users.
• Enterprise security / infosec teams adopting AI: They face a higher rate and cost of AI-related incidents and find traditional scanners noisy or ineffective for LLM-specific risks; they need tests that surface real, exploitable problems with proof YC profile.
• Fast-moving startups that must ship quickly: Manual pentests are slow and expensive; they need repeatable, continuous checks integrated into development to avoid introducing new AI risks on each release Casco site.
• Compliance, risk, and audit teams at regulated companies: They need reproducible evidence, clear reproduction steps, and remediation guidance to satisfy auditors and frameworks (e.g., SOC 2, NIST), not vague or noisy findings third‑party writeup.
• Platform/DevOps engineers running multi‑agent systems or third‑party tool integrations: Complex agent/tool chains expand the attack surface (cross‑tenant data leaks, infrastructure takeovers, RCE); they need threat scenarios that mirror these real failure modes to harden systems before incidents occur.

How would they acquire their first 10, 50, and 100 customers

• First 10: Founder-led pilots with teams building AI agents: run short hands-on engagements that uncover a few high‑impact, reproducible hijacks, deliver remediation steps, and convert pilots into references Casco site founder note.
• First 50: Publish technical write‑ups and reproducible tests; ship low‑friction CI/Slack/GitHub integrations so engineers can run checks with minimal onboarding; co‑sell/white‑label through nimble security consultancies as a faster alternative to manual pentests Casco site.
• First 100: Move upmarket with compliance‑oriented pilots (audit‑ready evidence and remediation playbooks for SOC 2/NIST), formalize MSSP/platform partnerships to reach infosec buyers, and run targeted ABM into risk/compliance leaders using early customer references YC profile third‑party writeup.

What is the rough total addressable market

Top-down context:

Casco straddles AI‑security and pentesting. Public estimates peg AI in cybersecurity around ~$25.4B in 2024 Grand View and penetration testing at ~$2.4–2.7B in 2024 Fortune Business Insights. The wider security software/spend context is ~$95B/$193B in 2024, respectively Gartner.

Bottom-up calculation:

Illustrative SAM for Casco’s current product: assume ~35k orgs actively building or deploying GenAI apps in the next year, with a median annual AI‑app testing budget of ~$60k (mix of autonomous testing and lightweight red‑teaming). That implies roughly ~$2.1B in near‑term spend that a purpose‑built AI testing tool could address, expanding as adoption and per‑org budgets rise.

Assumptions:

• Overlap exists between AI‑security and pentesting markets; top‑down figures are not additive.
• Active GenAI adopters (tens of thousands globally) will allocate a discrete line item to testing/evals as incidents and compliance pressures grow YC profile.
• Median per‑org testing budgets start modest (~$25k–$150k range) and expand with maturity and regulatory requirements.

Who are some of their notable competitors

• Robust Intelligence: Automated AI stress testing and validation to uncover model and system failures before deployment; overlaps with Casco on automated adversarial testing for AI systems.
• Protect AI: MLSecOps platform focused on securing AI/ML systems (scanning, risk management, and security tooling); competes for AI security budgets and assessments.
• CalypsoAI: AI security platform for GenAI governance, assessments, and guardrails; offers evals/red‑teaming services that overlap with AI app testing.
• Prompt Security: GenAI security platform with scanning and red‑team style tests for LLM apps; targets prompt injection and data exfiltration risks similar to Casco’s focus.
• Synack: Pentesting-as-a-service and red teaming platform; not AI‑specific but competes for pentesting budgets and continuous testing programs that Casco aims to replace or augment.