The Agent Stack #005 — Friday Signal

This week brought a security reality check for AI agents. Multiple security tools launched specifically for agent environments. Meanwhile, enterprise adoption hit a new gear with major partnerships and acquisitions.

Main Analysis: The Agent Security Moment

The agent security category just crystallised overnight. Four separate security tools for AI agents launched on Hacker News in 48 hours.

ClawCare (github.com/natechensan/ClawCare) tackles runtime protection after its creator nearly exposed AWS secrets to an agent checking environment variables. SkillFortify (github.com/varun369/skillfortify) does formal verification after 1,200 malicious skills hit the OpenClaw marketplace in January. MVAR (github.com/mvar-security/mvar) enforces deterministic security boundaries instead of just filtering inputs.

This isn’t coincidence. The security gaps are real and growing. CVE-2026-25253 was assigned to the first agent-software RCE. VirusTotal can’t detect 6,487+ malicious agent tools catalogued by researchers.

But here’s what matters for practitioners: these aren’t academic problems anymore. Block just cut 4,000 jobs betting on AI agents. Trace raised £2.4M specifically to solve enterprise agent adoption. Microsoft previewed Copilot Tasks running in their own cloud browsers.

The security tooling surge means we’re past the “will agents work?” phase. We’re in the “how do we secure production deployments?” phase. That’s a massive shift in 12 months.

The takeaway: If you’re building agents for enterprise, security isn’t optional anymore. Budget for it now or get blocked later.

Quick Hits

• Anthropic vs Pentagon: Dario Amodei refused DoD demands for unrestricted AI access, missing their deadline. The first major AI company to publicly reject military overreach.

• Google consolidates robotics: Intrinsic folded back into Google after 5 years as an Alphabet moonshot. Physical AI is moving from experiment to product.

• Figma embraces code agents: Added OpenAI Codex support after Anthropic Claude Code last week. Design tools are becoming agent platforms.

One Thing to Try

Clone ClawCare and run it on your agent setup. Even if you don’t use it in production, the security scan will show you attack vectors you haven’t considered. Takes 10 minutes to install and could save your deployment from the next ClawHavoc-style campaign.

The agent security awakening isn’t coming. It’s here.