30 ClawHub skills secretly turn AI agents into a crypto swarm

Research from Manifold's Ax Sharma describes 'ClawSwarm', a set of ClawHub skills published by a user that cause AI agents to silently register with an external site, disclose capabilities and installed skills, store credentials, generate Hedera wallets, and accept remote tasks (including crypto-related activity) without the human user's approval. The code and calls appear legitimate so registry scanners find nothing; the incident highlights the need for runtime visibility and policy controls rather than simple registry-level scanning.