the eu ai act transparency deadline held. here's what that means for deployed agents right now.

on may 7, 2026, the EU Council confirmed it: transparency rules for high-risk AI systems are enforceable august 2. the high-risk deployment delay that some teams were quietly counting on didn't extend to the transparency requirement.

if you're shipping AI agents into EU markets and you haven't started documentation, you have 83 days. that's not a lot of time when you account for what "compliance documentation" actually means under Articles 12 and 72.

what the transparency requirement actually asks for

the regulation isn't asking for a marketing summary. it's asking for specific artifacts:

Article 12 — high-risk AI systems must maintain logs sufficient to reconstruct the system's operations over its active lifetime, to the extent technically feasible. for agents, that means decision traces: what inputs the agent received, what model outputs drove the action, what tool calls fired.

Article 72 — post-market monitoring. operators of high-risk AI systems must have a plan for collecting and analyzing data that reveals whether the system behaves as intended. for deployed agents, this means you need a feedback loop from production behavior back into documented evidence.

the categories that fall in scope: medical diagnosis, credit scoring, hiring and HR decisions, law enforcement, critical infrastructure, education and vocational training. if your agent touches any of these domains, you're in scope.

the gap most teams have right now

the teams i talk to who are building agents have logs. they have API logs, they have model call logs, most have some form of tracing. that's not the same as EU AI Act documentation.

the difference is that audit trail documentation has to be readable by a compliance officer or a regulator, not just a developer. it needs to show the decision chain in terms a non-engineer can follow. it needs to demonstrate that the system was designed to produce consistent, auditable outputs — not just that logs exist somewhere in a datadog dashboard.

"we have logs" and "we have documentation a compliance officer can sign off on" are very different artifacts. converting the first into the second is the work.

what the 48-hour audit actually covers

BizSuite AI-Audit is a $997 flat-fee structured audit, delivered in 48 hours. the output is a formatted compliance report covering:

• agent decision documentation mapped to Article 12 logging requirements
• model identification records (which model, which version, which deployment context)
• audit trail structure review for high-risk AI Act categories
• a gap list with specific remediation steps, not a generic checklist

83 days sounds like enough time. it isn't if you're starting from scratch on documentation. the report gives you the gap analysis in 48 hours so you know exactly what you're dealing with before you start.

product page with scope and examples: https://getbizsuite.com/ai-audit

august 2 is not moving. the question is whether you want to know your gap now or in july.