AI agents nearly tripled in a year at companies, but liability for the decision remains human
A survey covered by TI Inside shows that operational autonomy has scaled to the systems, but legal responsibility for hiring, evaluating and dismissing with AI support still rests with the employer, with or without specific AI regulation in Brazil.
In focus: according to a report written by Walter Calza Neto and published by TI Inside on September 22, 2026, the adoption of AI agents at companies nearly tripled in just over a year: from an average of 5 agents in production per company in February 2025 to 13 in April 2026. The time needed to put a new agent into operation dropped from 4 days to less than 2. In the financial sector, institutional adoption of agents is projected to jump from 6% to 44% a year. The same report cites an estimate that up to 40% of corporate agentic-AI initiatives could be canceled by 2027 for lack of governance.
The central argument of the piece is a tension we had already been tracking from other angles in this column: operational autonomy scales to the systems, but legal responsibility remains human. The risk changes in nature once the agent stops merely responding to commands and starts acting on its own: accessing HR, finance and e-mail systems, and executing closings and decisions without human review of each transaction. When an agent decides on hiring, performance evaluation or dismissal, case law treats that algorithmic decision as the employer's own decision, with full liability attached to it.
The report also places the scenario in regulatory context: in the European Union, the AI Act's transparency obligations for public-facing systems have applied since August 2, 2026, and high-risk systems (which is exactly the category covering hiring, evaluation and dismissal) have until December 2027 to comply, under penalty of fines of up to 35 million euros or 7% of global revenue. In Brazil, Bill 2338/2023 (the AI regulatory framework) has been stalled since March 2025, which means that, for now, it is the LGPD (Brazil's data-protection law) and the Consumer Protection Code that already cover much of these situations, even without a specific AI law in force.
While Bill 2338/2023 stays stalled, the exposure of anyone already using AI agents to hire, evaluate or dismiss doesn't stay on hold either: it is already governed by the LGPD and the CLT (Brazil's labor code), with the employer answering for the agent's decision as if it were its own. Our Compliance, Audit & Outsourced-Workforce Management service includes mapping which automated decisions your company already delegates to AI systems, and whether anyone today is able to reconstruct and defend each one if it is questioned.
If an AI agent approved, evaluated or dismissed someone at your company tomorrow, would you be able to reconstruct and defend that decision?