Trust by design: Provenance for the agentic AI era

Hyperrealistic synthetic media, often called deepfakes, made content provenance — the record of where content came from and whether it changed — a public concern. 

Deepfakes eroded trust in what people could see, so the response focused on watermarking and artifact authenticity — tools that work well for static digital artifacts. But autonomous systems raise a different problem: verifying what a system did and under whose authority. It's a question authenticity alone cannot answer.

Agentic artificial intelligence marks this shift, introducing systems that can plan, use tools and act across multiple steps without a human directing each one. It moves the center of gravity from media authenticity to algorithmic accountability — from verifying the pixels in an image to verifying the chain of actions, tool use and delegated permissions that let an AI agent act on a human's behalf.

The agentic threshold: From authenticity to accountability

Authenticity concerns whether a digital object can be trusted as what it claims to be. Accountability is broader, encompassing whether a socio-technical system — the technology together with the people, policies and oversight surrounding it — can be understood, interrogated and accounted for.

In the agentic era, organizations must contend with systems taking unintended autonomous actions, misusing tools or operating beyond their guardrails. The European Data Protection Supervisor describes simple AI agents as performing tasks within defined bounds, while agentic AI systems reason and plan, use external tools, retain persistent memory across sessions and coordinate multiple subagents toward longer-term goals. 

The same pattern is increasingly ordinary in professional settings. A workplace copilot that reads an inbox, drafts a response, schedules a meeting and updates a task tracker without anyone approving each step already exhibits agentic characteristics. The more autonomously such tools act, the harder it becomes to reconstruct why a particular action was taken.