Agentic AI is no longer theoretical – it’s operational, autonomous and already reshaping industries. But as these systems begin to act independently, the legal and regulatory frameworks needed to govern them are struggling to keep apace. This article explores how different sectors are already adopting the technology, as well outlining the legal, ethical and regulatory guardrails needed to keep it safe, compliant and purposeful.
Published 23 October 2025
Agentic AI: What you need to know
Agentic AI marks a shift from passive AI tools to AI that actively takes decisions. These systems don’t just respond, they act. They’re designed to pursue goals, connect with external tools and operate independently, often without human oversight.
In 2024, OpenAI, Google and Salesforce launched agents that take action on users’ behalf. The promise? Smarter, faster outcomes. The risk? Systems that run indefinitely, making decisions we may not see, or control.
Jonathan Zittrain, writing in The Atlantic, highlights three traits that define agentic AI:
- autonomy: Agents pursue high-level goals without step-by-step instructions
- action: They interact with the wider world, not just with users
- persistence: They can run indefinitely, raising questions about accountability and control
Governance frameworks like the EU AI Act weren’t built for this, and may not offer adequate regulation or protection. Legal teams must now ask: how do we regulate AI agents that think, act and evolve on their own?
How agentic AI is being implemented across industries
Consumer and leisure: Adoption at speed
Agentic AI is transforming how consumer-facing industries operate. Retailers are using autonomous agents to personalise customer journeys in real time, while travel and hospitality firms are seeing a 133% monthly growth in AI agent actions. Players within the consumer and leisure sector aren’t just experimenting: they’re embedding agentic AI into core operations. For customers, this can mean faster service, smarter logistics and measurable ROI.
Financial services: Automating complexity
In banking, financial services, and insurance, agentic AI is being used to orchestrate multistep workflows. The goal? Sustainable performance gains. To achieve that, intelligent AI agents now deliver real-time, personalised investment advice, scanning markets and portfolios to flag opportunities or risks before clients even ask. In fraud detection, meanwhile, AI agents act autonomously in spotting anomalies, freezing accounts and alerting customers in seconds. And innovation is being embedded at the regulatory level too: the FCA’s “Supercharged Sandbox”, launched with NVIDIA, gives startups a safe space to test AI under live conditions.
AI and energy: From smart grids to billion-pound growth
Agentic AI is reshaping the energy industry from the inside out, driving smarter grids, predictive maintenance and real-time market participation. Across Europe, providers are using autonomous AI agents to reduce outages and reroute power instantly. The numbers speak for themselves: according to Market.US, the global agentic AI market in the energy sector is set to leap from $480m in 2024 to $10.7bn by 2034, with 36% of firms already onboard and reporting up to 30% savings.
Driving smarter vehicles and supply chains
In automotive and logistics agentic AI is advancing safety, personalisation and efficiency. In vehicles, it enhances in-cabin awareness, enables natural language interactions and streamlines design through data-driven simulations. In logistics, it automates customer support, optimises supply chains, and proactively manages compliance risks, boosting operational performance across both sectors.
Autonomy is here. Oversight is still pending
While the efficiency savings across sectors is impressive, it comes with risk. In many cases agentic AI systems will not wait for instructions. Instead, they act. That autonomy is powerful, but it may also be problematic. When AI agents operate indefinitely, connect with external systems and make decisions in real time, the risk isn’t just technical. It’s legal, ethical and reputational.
Guardrails must be built in, not bolted on.
With this autonomy comes a demand for embedded governance. That means designing systems with security from day one, auditing regularly, and aligning with evolving regulations like the General Data Protection Regulation (GDPR) and the EU AI Act.
Organisations must define internal boundaries, restrict access, and build oversight frameworks that evolve with the technology. Because the real risk isn’t what agentic AI can do. It’s what it might do, unsupervised.
Ethical considerations aren’t optional. They’re operational
Agentic AI also raises urgent ethical questions. Who decides what’s acceptable behaviour for an autonomous system? How do we monitor bias, protect data and ensure transparency? Businesses must move beyond intention to implementation, setting up governance boards, formal policies and real-time monitoring tools.
Global alignment is also key. International frameworks like the UN Guiding Principles on Business and Human Rights offer a foundation, but organisations must tailor ethical guardrails to their own risk profiles and stakeholder (as well as customer) expectations.
Regulators are catching up. Businesses must lead
The EU AI Act mandates human oversight for high-risk systems, but doesn’t yet address agentic AI directly. The UK’s approach to AI regulation generally is more cautious, offering best practices but no legislation. That leaves a gap – and a responsibility.
Organisations must fill that gap with proactive governance. The risk that agentic AI poses only underscores the need for companies to implement robust AI governance frameworks, policies and processes internally within their businesses now. Risk assessments, role-based permissions and API monitoring alone are no longer optional. Waiting for regulation isn’t a strategy. It’s a liability.
Final thought: Autonomy without accountability is a risk too far
Agentic AI is changing the game. But power without control is risk. The future of AI isn’t just about capability – it’s about the frameworks we build to keep that capability in check. With agentic AI, perhaps more than with any other type of AI system developed so far, human oversight and accountability remain key.