Clawdbot: 25K+ GitHub stars overnight. AI that actually executes tasks instead of just advising.
Runs locally. Persistent memory. Shell commands. Email management. Acts proactively through WhatsApp, Slack.
Filesystem, command execution, browser, email, smart home. 'No directory sandboxing', developer quote.
Traditional IAM manages humans. AI agents: persistent credentials, autonomous decisions, superhuman scale.
Ephemeral scoped credentials. Behavioral monitoring. Granular permissions. AI reasoning audit trails.
Self-hosted protects conversations from cloud storage, but API calls still transmit prompts to LLM providers.
Developers deploy AI agents without security oversight. Governance must evolve beyond prohibition.
Access scope. Credential management. Audit capabilities. Data handling. Isolation. Revocation speed.
OAuth-like protocols for AI agents. Compliance platforms. Machine identity tools. Liability frameworks.
Infrastructure built today determines whether the agentic future stays secure or descends into chaos.