August 2nd: The AI Act deadline your agents probably aren’t ready for

Did you know that August 2nd is the AI Act deadline? And are your agents ready for this? I'm Jamie Litherland, and I am the VP of Business Development. When the digital omnibus agreement landed in May, the headlines were reassuring. The AI act's high risk obligations have been pushed back. Boards exhaled. Compliance programs were quietly stood down. The assumption, entirely understandable and almost universally wrong, was that the whole regulation had been given more runway. It hasn't. Article 50 was not included in that postponement. It takes effect on August 2nd, 2026, exactly as written. And the fines that back it? Up to €15 million or 3% of global annual turnover, land on the same date. For most enterprises, this is the deadline that slipped through the gap. Article 50 is a specific part of the EU AI Act, not a global law. However, it can apply to businesses outside the EU if they develop AI tools used by people in the EU or if their AI generated content is directed at EU citizens. Article 50 is not about high risk AI. It is a horizontal transparency obligation that applies regardless of how your systems are classified. In plain terms, it covers four situations. 1. Conversational AI must tell users they are talking to a machine. 2. Generative AI outputs must be marked as artificially generated. 3. Emotion recognition systems must notify the people they are applied to. 4. Deepfakes must be labeled as such. The first of these is where most enterprise AI deployments sit, and it is where most enterprises are not yet ready. Over the last two years, AI agents have proliferated across the enterprise faster than anyone has tracked them. Customer service bots, HR assistants, IT help desk agents, onboarding flows. Many interact directly with employees, customers, or members of the public, which puts them squarely in scope of Article fifty. Here is the problem. Most organizations cannot answer the questions a regulator would ask. Questions like: Which of your AI agents interact directly with people? Do those agents disclose at first contact that they are AI? And can you evidence that disclosure? Not as a policy statement, but as a verifiable, auditable fact. If your answer to any of those is we think so or we'd need to check, that is not a compliance position. That is a liability. Most compliance programmes were built around the high risk classification framework. When the omnibus moved those deadlines, the programme moved with them. But Article 50 was never part of that framework. It rode along on the same calendar date by coincidence. When the high risk date shifted, Article 50 stayed exactly where it was. The agents most likely to trigger Article 50 are also the ones least likely to have gone through a formal governance process. They were built by a team, spun up in a business unit, deployed through a platform that made it easy not to think too hard. You cannot evidence disclosure obligations for agents you do not know you have. VOSS AI Agent Management gives you a single vendor neutral inventory of every AI agent in your estate, catalogued automatically from telemetry without manual registration, per agent activity visibility, and an audit trail of every interaction. Governance controls enforced consistently across your entire agent population regardless of platform. And when a regulator asks what your agents are doing, you have an answer. Not a best guess, but an evidenced, auditable one. With August 2nd approaching, that kind of visibility is worth having sooner rather than later, and it doesn't require a full re-platform to get there. Get in touch to find out more.