The EU AI Act marked a critical milestone this week, with rules for general-purpose AI models taking effect on August 2, 2025. These requirements include technical documentation, conformity assessments, and human oversight mechanisms to ensure systems operate safely and transparently. High-risk applications, such as hiring tools or public benefit systems, face stricter scrutiny to prevent bias and ensure accountability.

In the U.S., California introduced a state-level framework targeting agentic AI risks, emphasizing privacy protection, non-discrimination, and transparency. This contrasts with the Trump administration’s AI Action Plan, which prioritizes deregulation and unbiased AI principles through executive orders streamlining federal permitting and promoting “truth-seeking” AI systems. The administration also plans to review state-level regulations, potentially withholding federal funds from states with restrictive AI policies.

The Digital Fairness Act (DFA) extended its public consultation period to October 24, 2025, inviting feedback on regulating agentic AI. This signals growing global interest in addressing risks like data misuse and algorithmic bias, though specific proposals remain under debate.

Enterprises are adopting agentic AI protocols to standardize agent collaboration. Tools like LangGraph and AutoGen enable open-source frameworks for interoperable systems, while domain-specific agents (e.g., legal contract analysis) integrate specialized LLMs and custom workflows. These systems require AI observability tools to track decision chains and ensure compliance with frameworks like the NIST AI RMF.

Challenges persist, including cybersecurity threats from autonomous agents and lack of clear policies for high-risk AI applications. Employers face vicarious liability risks if agentic AI systems violate data privacy laws or exhibit bias, necessitating robust incident response plans and AI literacy training.