The race to deploy an AI workforce faces one important trust gap: What happens when an agent goes rogue?

Company leaders at Fortune’s Brainstorm AI panel said firms are in an arms race to deploy autonomous, agentic AI but adoption is uneven: Rubrik’s Dev Rishi mapped four phases—experimentation (≈50% of firms), formal production (25%), scaling (13%) and full autonomy (0%)—with many planning to move into production within two years. The dominant constraint is trust—security and governance are the No.1 blocker to shifting agents from knowledge retrieval to action—raising acute liability, compliance and reputational risks (Experian warned of likely breaches and tighter regulation) even as operators like Lowe’s report clear ROI from assistant agents and healthcare groups like Mass General Brigham urge human oversight. Panelists said practical requirements to accelerate adoption are enterprise-grade policy guardrails, enforceable procedures for failures, and clear identity, accountability, quality controls and post‑mortems, implying that vendors who can provide those controls and risk management will determine how quickly and safely agent deployments scale.

Analysis

Fortune’s Brainstorm AI panel highlighted uneven, early-stage enterprise adoption of agentic AI: Rubrik’s Dev Rishi categorized four phases with ~50% of 180 companies in experimentation, 25% formalizing prototypes into production, 13% scaling, 12% not started, and none at full autonomy. Firms in the experimentation cohort expect many to enter production within two years, but rapid adoption is contingent on solving trust issues.

Panelists identified security and governance as the primary blocker preventing agents from moving beyond knowledge retrieval to action; Experian explicitly warned of likely breaches, rogue agents, reputational losses and resultant regulatory pressure. That constraint is prompting conservative positions in regulated sectors and internal debates about whether to build or buy vendor agents.

Operational examples show both promise and limitations: Lowe’s reports strong, “tangible” ROI from agent companions across its 250,000 associates in large-format stores, while Mass General Brigham stresses mandatory human oversight in high-stakes care despite diagnostic upside in radiology. Practical prerequisites for scaling are concrete: systems that enforce policy guardrails, enforceable failure procedures, identity and accountability, output-quality checks, and post-mortem trails — requirements that will shape vendor leadership and enterprise pacing.

Key Decisions for Investors

• Prioritize exposure to software and infrastructure vendors that explicitly offer enterprise-grade governance, audit trails, identity/accountability frameworks and enforceable policy controls, because the article identifies these features as the gating factors for widescale agent adoption
• Be cautious on companies promising rapid agentic automation without documented security/governance frameworks, as Experian warned of breaches and regulatory backlash that would materially increase compliance and reputational risk
• Favor selective investment in operators demonstrating measurable ROI and thoughtful change management (the Lowe’s example), while requiring case-specific evidence of productivity gains and human-in-the-loop safeguards
• Monitor headlines and regulatory developments closely; a high-impact breach or explicit new regulation would be a catalyst to reprice risk for firms deploying agents at scale
• Reduce conviction in firms within highly regulated industries that plan to build bespoke agents without clear oversight mechanisms, and prefer those planning to buy vetted vendor solutions or who articulate robust failure-response policies