XTechAI CEO Warns of AI Risks as Systems Gain More Enterprise Permissions

Amadeo Pérez, CEO of XTechAI, has warned that as artificial intelligence (AI) moves from answering to acting, it needs stronger enterprise controls to prevent potential risks. Pérez, who founded XTechAI, believes that while AI can unlock significant productivity gains, it also poses new risks if companies grant AI systems excessive access to critical enterprise resources.

According to Pérez, the transition from traditional generative AI to AI agents that can perform a sequence of actions is a major shift. These agents can analyze inquiries, interact with customer relationship management (CRM) platforms, and prepare proposals, among other tasks. However, this transition raises concerns about the permissions and oversight required for AI systems.

Pérez advocates for the application of the principle of least privilege, a concept familiar to cybersecurity teams. This means that AI agents should receive only the permissions necessary to perform their specific roles. For example, a system that needs to read information should not have the ability to modify it, and an AI agent preparing an email should not automatically have the authority to send it. Sensitive operations should remain subject to human approval.

The CEO of XTechAI also emphasized the importance of human oversight in AI deployment. Companies should deploy different levels of autonomy based on the risk involved. Low-risk tasks can be completed automatically, while other actions can be prepared by an AI agent but require human validation before execution. Higher-risk operations, such as those involving sensitive data, financial systems, or important decisions, should remain under stronger supervision.

Pérez highlighted the need for clear records of what each AI agent has done, including the information accessed, tools used, and actions authorized. Logging, traceability, and recovery mechanisms will become increasingly important as AI gains operational access. He noted that the same AI technology that can help identify vulnerabilities can also be used to strengthen cybersecurity measures.

Amadeo Pérez also addressed the broader implications of AI on employment. He pointed to a report by U.S. Senator Bernie Sanders, which warned that AI, automation, and robotics could potentially replace nearly 100 million U.S. jobs over the coming decade. However, Pérez believes that technological transformation and labor disruption can coexist. Some tasks will disappear, some professions will change, and new roles will emerge. The central question, according to Pérez, is how workers and companies manage the transition.

Pérez supports a risk-based regulatory framework for AI, as seen in the EU AI Act, which aims to distinguish between different levels of risk. He believes that regulation should evolve alongside AI while avoiding unnecessary barriers to innovation. XTechAI's focus on practical use cases where AI can improve productivity, analysis, or automation while maintaining appropriate levels of security and human oversight reflects this approach.

As AI agents connect to real systems, the lines between AI and cybersecurity become increasingly blurred. A system that generates text creates one type of risk, while a system capable of interacting with credentials, databases, APIs, servers, and enterprise software creates another. The same AI technology that can help defenders identify vulnerabilities can also strengthen phishing campaigns, automate malicious activity, or produce fraudulent content at scale.

Pérez noted that recent warnings from inside the AI industry, such as those from former Anthropic researcher Jacob Coxon, who resigned in September 2026 while publicly raising concerns about the safety and control of increasingly capable AI systems, have added urgency to the debate. However, Pérez believes that such warnings should be taken seriously without automatically treating the most extreme scenarios as inevitable.

In conclusion, Amadeo Pérez of XTechAI emphasized the importance of strong controls and human oversight as AI moves from answering to acting. He highlighted the need for a risk-based approach to AI deployment and regulation, ensuring that the benefits of AI are realized while mitigating potential risks.