U.S. President Donald Trump has appointed a group of leading technology executives to a key advisory council that will help shape national policy on artificial intelligence and related technologies.

The appointments include Meta CEO Mark Zuckerberg, Nvidia CEO Jensen Huang, and Oracle Executive Chairman Larry Ellison, alongside other prominent figures such as Google co-founder Sergey Brin and AMD CEO Lisa Su. The group forms part of the President’s Council of Advisors on Science and Technology, known as PCAST.

The council is expected to play a central role in advising the administration on AI development, regulation, and global competitiveness. The White House said the panel could expand to as many as 24 members in the coming months.

Aligning Policy With Industry

The formation of the council reflects a broader effort by the administration to align closely with major technology companies as the United States seeks to maintain leadership in artificial intelligence.

Trump has positioned AI as a strategic priority during his second term, framing it as a critical area of competition with China. Shortly after taking office, he directed federal agencies to develop an AI Action Plan aimed at reducing regulatory barriers and accelerating innovation in the private sector.

The inclusion of senior executives from leading AI and semiconductor companies highlights the government’s reliance on industry expertise to guide policy decisions. Nvidia, for example, plays a central role in supplying the hardware that underpins AI systems, while companies like Meta and Google are advancing large-scale AI models and applications.

The council will be co-chaired by White House AI and crypto adviser David Sacks and technology policy official Michael Kratsios. Additional members from both industry and research sectors are expected to be announced.

Strategic Focus on AI Investment

The appointments come amid a surge in investment in AI infrastructure and development across the United States. Technology companies have committed significant capital to data centers, chips, and software systems as demand for AI capabilities grows.

The council’s work is likely to influence how the U.S. balances innovation with regulatory oversight, particularly in areas such as national security, economic competitiveness, and technological standards.

In addition to technology leaders, the council includes representatives from emerging sectors such as fusion energy, signaling a broader focus on advanced technologies that could shape future economic growth.

The move underscores the increasing importance of collaboration between government and industry in shaping the direction of AI development. As competition intensifies globally, policymakers are seeking to leverage private-sector expertise to maintain a technological edge.

The creation of the advisory council marks a step toward more coordinated national strategy in artificial intelligence, with input from some of the most influential figures in the technology sector.

A new study suggests that a widely used prompting technique in artificial intelligence may reduce performance in certain tasks, particularly coding and mathematics.

Researchers found that instructing AI models to adopt expert personas, such as telling them “you are an expert programmer,” can actually degrade accuracy. While the approach has become common in prompt engineering, the findings indicate that its effectiveness depends heavily on the type of task.

The research, conducted by academics affiliated with the University of Southern California, examined how persona-based prompting affects large language model performance. The study found that while such prompts can improve alignment and safety outcomes, they often harm factual accuracy.

Accuracy vs. Alignment Tradeoff

Persona-based prompting has been widely adopted since 2023, with users frequently framing requests by assigning roles to AI systems. The technique is intended to guide tone, structure, and behavior.

However, the study found that this method does not enhance factual reasoning. In benchmark tests using the Massive Multitask Language Understanding (MMLU) dataset, models prompted with expert personas performed хуже than baseline models. Accuracy dropped from 71.6% to 68.0% across multiple subject areas.

Researchers suggest that persona instructions may shift the model’s behavior toward following instructions rather than retrieving factual knowledge. In effect, the model focuses more on “acting” like an expert than accessing the information needed to answer correctly.

This effect was particularly evident in tasks dependent on pretraining data, such as coding and mathematics. In these areas, accuracy relies heavily on precise recall and reasoning, which persona prompts may disrupt.

Benefits for Safety and Structured Tasks

Despite the decline in accuracy, persona-based prompting showed benefits in alignment-focused scenarios. Tasks involving writing, role-playing, or safety constraints saw improved outcomes when models were guided by specific roles.

For example, assigning a “safety monitor” persona significantly increased refusal rates for unsafe or adversarial prompts. This suggests that personas can help enforce rules and behavioral constraints, even if they reduce factual performance.

The researchers concluded that prompting strategies should be tailored to the objective. For tasks requiring accuracy, minimal or direct prompts may be more effective. For tasks involving structure, tone, or compliance, detailed personas can provide advantages.

To address the tradeoff, the study introduces a method called PRISM, which dynamically applies persona-based behavior only when beneficial. The approach uses a gating mechanism to switch between standard model responses and persona-influenced outputs.

The findings highlight the complexity of prompt design in AI systems. As users increasingly rely on large language models for diverse tasks, understanding how different prompting techniques influence performance is becoming a critical area of research.

OpenAI is rolling out a redesigned shopping experience in ChatGPT, shifting its strategy toward product discovery after scaling back its earlier Instant Checkout feature.

The update allows users to search for products by describing what they need or uploading images for reference. ChatGPT then generates visual results that can be compared side by side, with details such as pricing, features, and reviews integrated into the interface.

OpenAI said the system has been improved in terms of speed, relevance, and product coverage, enabling more accurate and up-to-date results. The goal is to simplify a process that often requires users to navigate multiple websites and sources before making a decision.

Pivot Away From In-Chat Transactions

The redesign follows OpenAI’s decision to move away from Instant Checkout, a feature launched last year that enabled users to complete purchases directly within ChatGPT. The company initially positioned the tool as a key step toward AI-driven commerce.

However, the feature faced challenges. Analysts noted difficulties in onboarding merchants, maintaining accurate product data, and supporting common e-commerce functions such as multi-item carts and loyalty integrations.

OpenAI acknowledged these limitations, stating that the checkout model did not provide the flexibility required for a broad retail ecosystem. Instead, the company is now focusing on helping users discover products while allowing merchants to retain control over transactions.

Under the new approach, purchases are completed through external merchant platforms, often via in-app browsers or dedicated integrations.

Expanding Merchant Ecosystem

The updated experience is supported by deeper integration with retailers, who can now provide product feeds and promotional data directly to ChatGPT. This ensures their offerings are fully represented in search results.

Major brands including Target, Sephora, and Nordstrom have already adopted the new system. Additional integrations allow companies to build custom applications within ChatGPT, giving them more control over the user experience and transaction flow.

Walmart has introduced a dedicated in-ChatGPT shopping interface that supports account linking, loyalty programs, and payments. Meanwhile, Shopify is expanding its role by enabling merchants to connect storefronts to its catalog and complete purchases through embedded browsing experiences.

Shopify is also launching a new service called Agentic Plan, designed to help merchants without existing storefronts surface their products across AI platforms, including ChatGPT and Google Gemini.

The shift reflects a broader trend toward “agentic commerce,” where AI systems assist users in navigating complex purchasing decisions rather than handling transactions directly.

OpenAI’s updated strategy positions ChatGPT as a discovery layer within the retail ecosystem, focusing on intent generation and product comparison. As AI becomes more integrated into shopping workflows, the balance between platform control and merchant autonomy is likely to remain a key area of development.

Anthropic has introduced new capabilities for its Claude AI model that allow it to control a user’s computer and execute tasks autonomously, marking a significant step toward fully agentic systems.

The update enables users to assign tasks to Claude remotely, including from a mobile device, with the AI carrying out actions directly on a connected computer. According to the company, Claude can open applications, navigate web browsers, and interact with files such as spreadsheets.

In one demonstration, a user asked Claude to export a presentation as a PDF and attach it to a meeting invitation. The system completed the multi-step workflow without manual intervention, highlighting the model’s ability to operate across software environments.

Push Toward Always-On AI Agents

The release reflects a broader shift in the AI industry toward building agents capable of acting independently on behalf of users. These systems are designed to handle complex, multi-step tasks and remain active even when users are not directly engaged.

Anthropic’s update follows growing interest in agent-based tools such as OpenClaw, which gained traction for enabling users to interact with AI through messaging platforms like WhatsApp and Telegram. OpenClaw operates locally on devices, allowing access to files and applications, a model that Anthropic is now partially replicating.

Competition in this space is intensifying. Nvidia recently introduced an enterprise-focused version of OpenClaw, while OpenAI has recruited its creator as part of efforts to develop next-generation personal AI agents.

Anthropic’s new feature builds on its broader ecosystem, including tools like Dispatch, which allows continuous interaction with Claude across devices and supports task assignment in an ongoing workflow.

Safeguards and Early Limitations

Anthropic emphasized that the computer control capability is still in an early stage compared to Claude’s performance in coding and text-based tasks. The company acknowledged that the system can make mistakes and that security risks remain a concern.

To address these issues, Claude requires user permission before accessing new applications or executing certain actions. The company said it has implemented safeguards to reduce potential misuse, though it noted that threats continue to evolve.

The introduction of computer control highlights both the potential and the challenges of agentic AI. While the technology promises increased productivity and automation, it also raises questions about reliability, security, and user oversight.