What happened
At Assemble London (Oct 14), Chainguard introduced Harness: point Anthropic Claude Mythos 5.1 at a live, running copy of your application inside an isolated microVM (no egress, no credentials) to find logic flaws, injection, and resource exhaustion; every exploit is replayed to confirm, and fixes are verified. Beta waitlist opened Oct 14; will be purchasable via Claude Marketplace against existing Anthropic commitments.
Why it matters
Democratizes frontier-model vulnerability hunting that was previously reserved for a few labs, with a reproducible, human-reviewed deliverable — a notable entrant in AI-driven offensive/ASM and direct competition to AWS Security Agent (Continuum) and Unit 42 frontier-AI defense services.
Applicability
Product security/AppSec teams in orgs with Anthropic commitments should join the beta waitlist now to evaluate it as a complement to static scanner-led programs.