Prove there’s a human in the loop.
AI agents are getting the keys: payments, records, infrastructure, contracts. When an agent reaches a consequential action, Dapple inserts a biometric checkpoint — cryptographic proof that a specific, verified human approved what happens next.
Book a Demo Read: The EU AI Act & Human OversightAgent permission is the new attack surface
Enterprises are wiring AI agents into systems that move money and change records — and regulators have noticed. The question every security and compliance team now faces: when the agent acts, who approved it, and can you prove it?
EU AI Act — Article 14
High-risk AI systems must operate under effective human oversight, with obligations now slated for 2 December 2027. That’s a preparation window, not a reprieve — “a human clicked OK somewhere” won’t satisfy an auditor; a biometric-verified approval will.
AI governance policies
Boards are demanding control evidence for agentic AI. Dapple turns a policy line — “sensitive actions require human approval” — into an enforceable, logged, cryptographic control.
Agents can be spoofed
Prompt injection and stolen session tokens let attackers drive agents. A biometric gate means a hijacked agent still can’t execute the sensitive step — the human checkpoint doesn’t phish.
The rulebook is global
Human oversight of AI isn’t a European quirk — it’s converging guidance across the US federal government, international standards bodies, state legislatures, and the security community. Start here:
EU AI Act — Article 14
The one that starts the clock: high-risk AI systems must operate under effective human oversight, with obligations now slated for 2 December 2027. If you sell into or operate in the EU, the preparation window is open — and the EU expects implementation to already be underway.
Read the regulation at EUR-Lex →NIST AI Risk Management Framework
The de facto US standard for trustworthy AI. Its GOVERN and MANAGE functions call for defined human oversight roles and accountability for AI-driven decisions — the checklist US enterprises are audited against.
Read the framework at NIST →ISO/IEC 42001 — AI Management Systems
The first certifiable international standard for AI governance. Organizations pursuing certification must demonstrate controls over AI system use — including who authorizes consequential actions.
See the standard at ISO →OWASP GenAI Security Project
The security community’s canonical catalog of LLM and agentic-AI risks — excessive agency, prompt injection, and insecure agent permissions among them. Required reading for anyone gating agent actions.
Explore the OWASP guidance →Colorado AI Act (SB 24-205)
The first comprehensive US state AI law — passed in Dapple’s home state — requiring developers and deployers of high-risk AI to use reasonable care, disclose usage, and maintain risk management programs.
Read the bill at leg.colorado.gov →A checkpoint, not a bottleneck
Your AI agents move at machine speed — oversight takes a heartbeat. One glance, one touch, cryptographic proof a human said go. See it live in twenty minutes.
Book a DemoDesign partner program open
The biometric human-in-the-loop category is forming right now. We’re working with early enterprises deploying agentic AI in finance, operations, and security workflows — pricing is scoped per deployment. Claim a seat at the design table.
Contact Sales