Synthetic Media & Agents

An agent that can publish is a producer of synthetic media, and the obligations that attach to it are not access control. They are disclosure, provenance and consent — a different body of law and a different practice.

Agentic AI Security Program Kit — what the agent may do. 27 fields per agent resolving to a blast-radius tier, a guardrail standard with a gate on each action class, OWASP ASI threat modelling with 18 test cases, the logging requirements and the playbook for the hour after one goes wrong. Publish is its own gate, and it is the one that hands off to the other half of this pair.

Synthetic Media & Deepfake Defense Kit — what has to be true before anything goes out. Disclosure built once — a visible caption, the platform's native label, C2PA content credentials and a register row — satisfying EU Article 50, the Chinese, Korean and Indian labelling rules, the YouTube, TikTok and Meta policies, New York's synthetic-performer advertising rule and the state election statements together. Plus the likeness and voice releases the digital-replica statutes now require, and the Defend half for when your own executives are the ones being cloned.

Why the pair, and why now. Most agent programmes gate money and deletion and stop there, because those are the failures people picture. Publishing is the one that scales: an agent posting to a channel, generating a marketing asset or answering a customer in your brand's voice is producing content that a dozen regimes now require you to label, attribute and hold consent for — and it is producing it faster than a review queue can absorb. Gate it in the agent standard, discharge it in the disclosure practice, and the two register rows reconcile.

This pairing became real in AGENT-01 v1.1, which split Publish out of Send as its own action class and wrote the disclosure requirements into the gate. Buying the pair is buying both halves of that gate.