Dieser Artikel ist derzeit auf Englisch verfügbar.
AGI and the Board: What Directors Need to Understand About Artificial General Intelligence
Artificial General Intelligence is not a distant science fiction scenario — it is a strategic risk that boards of major technology companies, regulators, and governments are actively planning for. What AGI means, what the governance implications are, and what boards should be doing now.
Key Takeaways
AGI — AI systems with general reasoning capabilities comparable to or exceeding human intelligence across most domains — is actively being pursued by major labs. The governance implications arrive before AGI does.
The most important near-term AGI governance question is not 'when will AGI arrive' but 'what governance structures should be in place before capabilities significantly exceed current AI systems.' The answer is: now.
Frontier AI governance — the emerging framework for governing the most capable AI systems — is already being implemented by leading labs and referenced in government AI safety frameworks including the UK AI Safety Institute and the US AI Safety and Security Board.
The three governance actions boards should take now: understand what frontier AI systems your organisation uses or depends on, include AGI risk in enterprise risk management, and ensure your AI governance framework is designed to scale with capability increases.
Existential risk is not the only AGI governance concern — the governance of highly capable AI in the next 3-5 years (before AGI, if it arrives) is where the immediate practical and regulatory obligations lie.
"Nur zu Informationszwecken. Dieser Artikel stellt keine rechtliche, regulatorische, finanzielle oder professionelle Beratung dar. Konsultieren Sie einen qualifizierten Spezialisten für spezifische Beratung."
What AGI actually means — and what it does not
The term Artificial General Intelligence suffers from both overuse and imprecision. In its most rigorous definition, AGI refers to AI systems that can perform any intellectual task that a human can perform, with at least human-level capability, and can generalise from one domain to another — as humans do. Current AI systems, including the most capable large language models, are not AGI by this definition. They are extraordinarily capable in specific domains, fail in characteristic ways that humans do not, and cannot reliably transfer learning from one domain to another in the way human intelligence can.
The governance-relevant question is not whether AGI exists today but what the trajectory of capability development implies for governance planning. The organisations building frontier AI systems — OpenAI, Anthropic, Google DeepMind, and others — believe they are on trajectories that will produce significantly more capable systems within years, not decades. These organisations have published safety frameworks, deployed alignment research teams, and in some cases advocated for regulatory oversight precisely because they take this trajectory seriously. Boards should take it at least as seriously as the organisations building these systems do.
Frontier AI governance: what is already happening
The governance of the most capable current AI systems — called frontier AI — is already an active regulatory priority in major jurisdictions. The UK AI Safety Institute was established specifically to evaluate frontier AI systems for safety and security risks. The US AI Safety and Security Board advises on the safety of advanced AI. The EU AI Act's obligations for General Purpose AI Models with "systemic risk" — models trained with more than 10^25 FLOPs — create mandatory safety obligations for the most capable frontier systems. These frameworks are real, operating, and relevant to organisations that use or depend on frontier AI systems today.
The frontier AI governance frameworks converge on several common requirements: evaluation of AI systems for dangerous capabilities before deployment; disclosure of training methodologies and safety evaluations; incident reporting for safety-relevant events; and in some cases mandatory pre-deployment authorisation for the most capable systems. Organisations that use frontier AI APIs or build products on frontier AI models are exposed to these requirements through their vendor relationships and through the supply chain governance obligations of the EU AI Act.
What boards should do now
Three practical actions are appropriate for most boards, regardless of their organisation's direct involvement with frontier AI development. First, map your frontier AI exposure: which AI systems does your organisation use that are developed by frontier AI labs? What are the governance and safety commitments of those labs, and how do they affect your risk profile? Second, include AGI-scenario planning in enterprise risk management: not as prediction of specific outcomes, but as scenario analysis of what material capability increases would mean for your business model, competitive position, and regulatory obligations. Third, ensure your AI governance framework is designed to scale: governance frameworks built around current AI capabilities will require updating as capabilities advance — building in a review mechanism ensures governance remains current.
The nearer-term governance challenge: highly capable AI before AGI
The governance challenge that most immediately deserves board attention is not AGI itself but the AI systems that will exist in the next three to five years — systems significantly more capable than today's, but not necessarily meeting the definition of AGI. These systems will be able to autonomously complete complex multi-step tasks, generate synthetic content indistinguishable from expert human output, and be deployed in agentic configurations where they take sequences of consequential actions with limited human oversight. The governance frameworks for these near-term systems are the practical priority. The EU AI Act, the NIST AI RMF, and ISO 42001 were designed with current AI in mind and will require updating as capabilities advance — being ahead of this governance evolution, rather than reactive to it, is the strategic advantage.