We deployed PromptKing against our own Slack workspace. Every AI agent session ran without human authorization. Here is what the data shows — and why it matters for EU AI Act Article 14.
Slack Governance · Tagged vs Ambient — the governance split that matters
#pk-engineering generated the most untagged ambient activity this week. Ambient agents in this channel have operated without per-session human authorization.A tagged session is one where a human explicitly invoked an AI agent — an @mention, a slash command, a deliberate trigger. A human was present. A human made a decision.
An ambient session is everything else. The agent acted. No human triggered it. No human authorized that specific action. It happened in the background, in your Slack channels, right now, and your compliance team has no record of it.
EU AI Act Article 14 requires human oversight of high-risk AI systems. It does not say “humans must be able to intervene.” It says humans must be in a position to intervene — which requires knowing what the system is doing before it does it.
When 100% of your AI agent sessions are ambient, your compliance team cannot satisfy Article 14. There is no per-session human authorization record. There is no audit trail that demonstrates oversight. There is only activity — unattributed, unreviewed, and ungoverned.
“We found this in our own workspace. Not a client's. Ours. That is how we know the problem is real — because we were the problem.”
PromptKing's Slack Governance connector classifies every agent session as it happens. It reads event metadata — not message content, never prompt text — and applies a deterministic tag: tagged (human-initiated) or ambient (no human trigger detected).
When the ambient percentage climbs — as it did in #pk-engineering — a drift signal fires. The channel is flagged. The compliance record is written. The governance layer has evidence that Article 14 oversight was either present or absent for every session.
That record is exportable as an Evidence Pack. It names the channel, the session count, the ambient percentage, and the drift classification. It is the document your compliance team needs when the regulator asks.
The total session count is less important than the split. An enterprise running 10,000 AI sessions per week with 40% tagged has a governance story. An enterprise running 10,000 sessions with 0% tagged has an Article 14 problem — at scale.
PromptKing tracks this ratio in real time across every connected vendor — Slack, Microsoft 365 Copilot, GitHub Copilot, Google Gemini, AWS Bedrock, and IBM Watsonx. The governance split is the metric your compliance team needs to report on. We make it visible.
PromptKing connects to your existing AI subscriptions. No agents installed. No prompt access. Governance from the billing and event layer — the only place that sees everything.
See Your Governance Split →