As artificial intelligence evolves from chatbots to autonomous "agents" capable of executing complex tasks without human oversight, a leading AI ethics expert is sounding the alarm that most organizations lack the infrastructure to manage the mounting risks.

Dr. Reid Blackman, founder of AI consultancy Virtue and author of "Ethical Machines," warns in a new Harvard Business Review analysis that companies are approaching a "head-spinning quagmire of incalculable risk" as AI systems become increasingly sophisticated and interconnected.

The warning comes as businesses rush to deploy AI agents - systems that can perform sequences of tasks, make decisions, and interact with other AI systems - without adequate safeguards in place.

The Complexity Curve

Blackman outlines five stages of AI complexity that organizations navigate, from basic predictive tools to interconnected networks of AI agents that can communicate across company boundaries. Most companies, he says, built their AI risk programs around simple, narrow AI applications and haven't scaled their oversight capabilities to match the technology's rapid evolution.

"In my work helping Fortune 500 companies design AI ethical risk programs, I have yet to encounter an organization that has the internal resources or trained personnel to handle Stage 2, let alone the later stages," Blackman writes.

The stages progress from basic AI tools to what he calls "multi-model multi-agentic AI"—systems where multiple AI agents within and between organizations can interact, make decisions, and take actions at speeds far beyond human comprehension.

Breaking Point for Human Oversight

A critical concern is the breakdown of traditional "human-in-the-loop" oversight. While earlier AI systems generated outputs that humans could review before acting, advanced AI agents operate too quickly and process too much information for meaningful human intervention.

"There's just too much data for any human to possibly process in real time," Blackman explains, emphasizing that this shift places "enormous weight" on pre-deployment testing and real-time monitoring systems that most organizations lack.

The analysis highlights a significant employee training deficit. Unlike traditional AI that required mainly data science expertise, generative and agentic AI systems require widespread organizational literacy. Employees across departments need training not just in using AI tools responsibly, but in recognizing when systems malfunction.

"The most successful companies with whom I've worked have one thing in common: They invested and continue to invest heavily in employee training before deploying the technology, not after problems emerge," Blackman notes.

The risks extend beyond technical glitches to potential "business and brand-defining disasters." As AI agents gain the ability to take digital actions - such as conducting financial transactions - and communicate with external systems, a malfunction or misalignment could cascade rapidly across networks.

Blackman frames this challenge around what he calls "The Ethical Nightmare Challenge," which asks leaders to identify potential ethical disasters from AI use, create resources to prevent them, and train employees accordingly.

A Path Forward

Despite the stark warnings, Blackman offers a roadmap for organizations. Rather than attempting to solve all AI risks at once, companies should honestly assess their current position on the complexity curve and build appropriate capabilities before advancing to more sophisticated systems.

Key recommendations include:

• Comprehensive employee training programs that go far beyond typical compliance videos
• Robust realtime monitoring systems
• Clear intervention protocols for when AI systems malfunction
• Rigorous pre-deployment testing frameworks

The analysis comes as major technology companies like Microsoft, Google, Anthropic, and OpenAI continue developing increasingly powerful AI systems, while regulatory frameworks struggle to keep pace with technological advancement.

Organizations face what Blackman characterizes as "a stark choice" between proactively building proper infrastructure now or waiting for a catastrophic failure to force action—likely at much greater cost and with significant damage to stakeholder relationships.

Read the original article here.