Samuel Thompson

Stanford’s AI Index Report reveals a narrowing US-China gap in AI model performance, with China showing increased publication and patent volume. AI safety benchmarking significantly lags behind capability assessments, leading to rising incidents and organizational governance struggles. Public anxiety about AI’s impact grows, contrasting with expert optimism, and the US shows low trust in its government’s ability to regulate AI responsibly.

SAP is integrating agentic AI into its SuccessFactors HCM suite, starting with the 1H 2026 release. This move aims to proactively resolve workflow bottlenecks and enhance operational intelligence across HR functions like recruitment and payroll. The AI agents will monitor systems, identify anomalies, and offer context-aware solutions. This technology tackles data synchronization issues, automates troubleshooting, and simplifies knowledge retrieval for employees. SAP’s approach also streamlines onboarding by integrating candidate data seamlessly and offers a new extensibility wizard for custom development on its Business Technology Platform. Furthermore, the release bolsters compliance with pay transparency insights and strengthens talent management through enhanced skills governance.

Scotiabank launched “Scotia Intelligence,” a unified AI framework with strong governance. It aims to empower employees with AI tools while adhering to strict ethics and security. The initiative streamlines operations, enhances customer experience through predictive prompts, and accelerates software development with automated coding. Demonstrating significant ROI through efficiency gains and improved client interactions, Scotiabank emphasizes responsible AI deployment with comprehensive training and rigorous review processes.

Hyundai Motor Group is pivoting beyond automotive to “physical AI,” integrating intelligence into robots for physical world interaction. With a $26 billion U.S. investment, the group aims for human-robot collaboration, scaling humanoid robot production for factories and exploring applications in logistics and mobility. This strategy, driven by Chairman Chung Eui-sun, emphasizes synergy between humans and machines to enhance efficiency and quality, complemented by significant investment in hydrogen technology.

The rise of powerful, locally executable AI models like Google’s Gemma 4 challenges traditional enterprise security. These models can run on edge devices, bypassing traditional perimeter defenses and making network traffic monitoring ineffective for offline AI processing. This creates blind spots for security, especially in regulated industries requiring auditability. CISOs must shift focus from blocking models to controlling intent and system access, as identity platforms become the new firewall. Enterprises need new endpoint detection tools to monitor local AI inference and adapt security policies to acknowledge that compute execution is no longer solely cloud-based.

Businesses are adopting a cautious approach to AI, favoring tools that augment human decision-making over full automation. Sectors with high financial or legal risks prioritize AI that is manageable, verifiable, and trustworthy. S&P Global Market Intelligence exemplifies this by integrating AI into its Capital IQ Pro platform to assist financial analysts, ensuring AI outputs are tied to verified sources. While AI adoption is widespread, many organizations use it for tasks like summarization, not independent action. Trust in AI systems relies on their ability to explain reasoning, cite sources, and operate within defined parameters.

To protect margins and foster innovation, businesses must prioritize robust AI governance, moving from opaque proprietary systems to open infrastructure. As AI becomes foundational, closed development becomes untenable due to complexity, security risks, and integration challenges. Open-source AI enhances operational resilience through broad scrutiny and collaborative improvement. Embracing transparency and open foundations is crucial for enterprise AI’s future, enabling adaptability, innovation, and public legitimacy.

Next-generation AI assistants, from Apple and Qualcomm, will offer advanced capabilities for task management and app navigation. However, development prioritizes user control and security, incorporating explicit approval checkpoints for sensitive actions like financial transactions. This “human-in-the-loop” model limits AI autonomy, ensuring users retain final decision-making authority and data privacy through granular access controls and integration with secure partner services. This controlled approach aims to mitigate risks and foster trust in agentic AI.

Meta’s Muse Spark marks a significant shift from its open-source Llama initiative. This proprietary, multimodal AI model, developed after a $14.3 billion investment, surpasses Llama 4 in benchmarks, particularly in healthcare. Unlike its predecessors, Muse Spark’s advanced capabilities are not freely available. This move, while driving Meta’s stock up and targeting billions of users directly, has generated skepticism within the developer community awaiting future open-source releases.

To mitigate risks in AI agentic systems, organizations must prioritize robust identity protocols, auditable logs, policy enforcement, human oversight, rapid revocation, and thorough documentation. Employing tools like SDKs that cryptographically sign actions, similar to blockchain, creates an immutable audit trail. A centralized, verbose system of record for all agentic AI activities is crucial for governance, surpassing scattered text logs. Maintaining an “agentic asset list” with unique identifiers, capabilities, and permissions is essential, aligning with regulations like the EU AI Act’s mandates for ongoing, evidence-based risk management and interpretable AI systems.