Mitsotakis admits no government is ready for what AI brings

"We're already fighting yesterday's battle." With that sentence, Greek Prime Minister Kyriakos Mitsotakis summed up in an interview with TechCrunch something few leaders say out loud on a trade trip: no government is ready for what artificial intelligence is about to do.

The conversation was published on 22 September, in the middle of a trade mission. In that format the usual move is to sell the country as an investment destination, repeat growth figures and avoid anything that sounds like doubt. Mitsotakis did the commercial part, but according to the TechCrunch reporter, he also acknowledged that states are not keeping up.

Why the sentence carries more weight than it seems

Political speeches about AI usually swing between two poles: the promise of productivity and the warning about risks. What Mitsotakis raises is something else, a problem of pace. Legislative and administrative cycles are measured in years; model releases, in months.

Anyone who works with this technology every day notices it. So far in 2026 we have gone through Claude Sonnet 4.6, Claude Opus 4.8 with an optional 1M token context window and, more recently, Claude Fable 5. Each jump changes what is reasonable to automate: tasks that a year ago required constant supervision are now delegated to agents that chain tools through MCP servers. A rule drafted with question-answering chatbots in mind falls short when software already takes actions on its own.

That this reading comes from a European Union head of government, and not from a lab or an investor, is what makes it relevant.

Europe, between the rules and practice

The EU approved the AI Act in 2024, with a phased rollout that started with prohibited practices and continued with obligations for general-purpose models from August 2025. On paper, Europe is the region with the most detailed framework. In practice, companies are still asking how that translates to their specific case, and the national authorities in charge of supervision are still building technical capacity.

Greece, moreover, is not starting from scratch on public-sector AI: in 2025 it signed an agreement with OpenAI to bring generative AI tools into the education system. In other words, the prime minister admitting that governments are late is also one of those who moved early to adopt the technology. That combination, adopting fast and regulating late, describes the situation of many European countries quite well.

Who this reading is useful for

The interview is not a regulatory announcement, but it is a signal worth interpreting depending on the profile.

Companies integrating LLMs into sensitive processes: if a head of government admits regulation lags behind, the likely outcome is that rules arrive later and all at once. Documenting now what each agent does, with which data and under which permissions, lowers the cost of adapting later.

Public administrations: the implicit message is that waiting for a finished framework before building internal teams does not work. The technical capacity to evaluate systems has to be built in parallel.

Technical teams: a good part of governance can be solved in the architecture itself. In Claude Code, for example, PreToolUse hooks can block or log actions before an agent executes them, and limiting which MCP servers each subagent can access narrows the scope of errors.

What remains open

Acknowledging the lag is the first step, but it does not answer the practical question: what should a government do when it knows its tools arrive late. The options discussed in Europe range from strengthening supervisory authorities with technical profiles to relying on codes of conduct that can be updated faster than a law. None has been tested at scale.

Meanwhile, responsibility shifts toward whoever deploys the systems. If the rules do not draw the line precisely, contracts, internal audits and the judgement of the teams putting agents into production will.

At ElephantPink we appreciate Mitsotakis's candour, because it describes what we see in projects: the technology moves faster than any external framework. That is why we recommend not waiting for regulation to set clear limits on agents; doing it by design is cheaper than doing it afterwards.