When Tolmo published its working security checklist for AI-native products last week on Hacker News, it landed with a thud โ€” just two points and zero comments at time of writing. Which is a shame, because the thing is genuinely useful if you're an engineering leader trying to figure out what secure actually looks like across your company's growth stages. The checklist maps six core security categories: Application Security, Cloud & Infrastructure, Business Logic & Access, Attack Surface, Sensitive Data & Secrets, and Monitoring & Response. Each category stacks progressive requirements from Seed stage all the way through Series C, meaning every item at later stages includes everything before it โ€” a clean mental model for teams that don't want to juggle multiple maturity frameworks.

Application Security

On the app side, Tolmo starts brutally simple: version control with mandatory code review on every change and dependency scanning in CI. That's Seed-level table stakes. By Series A, you're running continuous penetration testing against production looking for real exploitable vulns. Series B is where it gets spicy โ€” red-teaming agent workflows specifically for jailbreaks and data exfiltration before anything ships. If you're building AI-native products, that's not paranoia; it's Tuesday.

Cloud & Infrastructure

Cloud security follows a similar arc: encryption at rest and in transit by default, least-privilege IAM with prompt offboarding at Seed, then infrastructure-as-code with policy-as-code checks in CI by Series A. By Series B, you're actively mapping privilege-escalation and lateral-movement paths across your accounts โ€” which is exactly the kind of attacker mindset most startups don't adopt until after they've been burned.

Sensitive Data & Secrets

The Sensitive Data & Secrets section hits hardest for AI companies specifically. It calls out data retention and PII policy requirements not just for your systems, but for AI model inputs and outputs โ€” a gap that trips up a shocking number of early-stage teams. The checklist also demands validation of leaked credentials on exposure and rotation workflows, because stuffing API keys in source control is still happening at Seed companies with alarming regularity.

Monitoring & Response

Monitoring starts with centralized logs and alerting on high-risk events (Seed), progresses to ingesting telemetry from tools like Datadog, Splunk, or Wiz by Series A, then adds 0-day disclosure monitoring with named owners and SLAs at Series B. By Series C, you're achieving SOC 2 / ISO 27001 and reporting security posture directly to the board โ€” which is increasingly a fundraising and enterprise sales requirement anyway.

Key Takeaways

  • Security requirements are organized by company stage so you only focus on what's relevant now
  • AI-specific concerns like agent workflow jailbreaks and model I/O data policies get explicit treatment
  • Red-teaming for jailbreaks before launch is flagged at Series B โ€” earlier than most teams think to do it
  • SOC 2 / ISO 27001 appears as a Series C milestone, reflecting enterprise sales realities

The Bottom Line

This checklist won't replace a dedicated security team, but it's the best sanity-check I've seen for solo CTOs or small engineering orgs trying to ship securely without drowning in framework documents. Bookmark it, work backward from where you want to be at Series B, and start automating what Tolmo's agents handle โ€” because if you're not red-teaming your AI agent workflows before launch, someone else will find the gaps for you.