Today's Flow

As the pace of AI technology advancement accelerates, major companies are focusing on efforts to overhaul internal safety monitoring systems within their research labs and maximize infrastructure efficiency. The most prominent change is Anthropic and OpenAI’s push to station independent safety evaluators within their own research facilities. External researchers While positively evaluating the expansion of internal accessibility, it is pointed out that transparency and independence, along with institutional regulation, must be in place for effective oversight to occur. This reflects the demands from within and outside the industry to ensure the effectiveness of safety verification even amid the race for development speed.

In the hardware and infrastructure domain, NVIDIA-led improvements in performance and power efficiency stand out. NVIDIA has raised the bar for inference economics by achieving benchmark-leading performance with its next-generation Vera Rubin NVL72, and has formed an alliance with Google and Emerald AI to manage data center power through grid integration. Furthermore, in the software and services ecosystem, Microsoft is expanding its educational collaborations, GitHub is enhancing Copilot’s budget management and security scanning capabilities, and OpenAI is introducing marketing agents tailored for older adults, among other developments, broadening practical application areas.

Key News

Topic: Anthropic and OpenAI's Push for Resident Safety Evaluators at Research Institutes and the Challenge of Independence

Anthropic and OpenAI are pursuing plans to station independent safety evaluators within their own AI research institutes. While researchers welcome the unprecedented level of access to the institutes, they emphasize that effective oversight and supervision require high transparency and robust guarantees of independence, and that long-term institutional frameworks are essential. It warns that linkage with norms is necessary. It highlights that securing the transparency of evaluation results and independent decision-making authority, rather than merely relying on internal placement, is the key to ensuring safety.

Topic: Anthropic integrates Claude chat window and Co-Work collaboration features into a single interface

Anthropic has integrated Claude’s conversational interface and its collaborative feature, Cowork, into a single unified interface. This integrated functionality is being rolled out sequentially to subscribers of the paid Pro and Max plans. By consolidating conversation and collaboration features in one place, it reduces the friction in users’ workflows and boosts productivity. This is an action to raise it.

Topic: NVIDIA Vera Rubin NVL72 Sets Leading Performance Record in MLPerf Inference v6.1

NVIDIA's Vera Rubin NVL72 achieved top-tier performance in its debut on the MLPerf Inference v6.1 benchmark. NVIDIA emphasized that system performance, efficient infrastructure scalability, and continuous software optimization are key drivers determining the economics of AI inference. High system performance increases value by generating more tokens, Through efficient scaling, it is possible to proportionally increase processing capacity when adding hardware, thereby reducing the operational costs of large-scale services.

Topic: Emerald AI, Google, and NVIDIA Launch the AEMA Alliance for Flexible Power Management

Emerald AI, Google, and NVIDIA have officially announced the formation of the "AI Energy Management Alliance (AEMA)," a consortium aimed at dynamically managing power consumption in AI data centers. This alliance was established with the recognition that the responsible expansion of AI data centers must be closely linked not only to internal technological innovation but also to broader innovations across the external power grid. It is the first collaborative initiative launched by the industry.

Topic: General Availability of GitHub Copilot Budget Increase Request Feature and Enhanced Convenience for Security Scans

GitHub has transitioned the approval process for team members to request additional budget when Copilot AI credits are exhausted to General Availability (GA). Previously, access to related features was immediately blocked once all provided credits were used; however, the new flow enables smoother requests for budget increases. Additionally, it supports AI scanning in code scanning. We have improved functionality to broaden accessibility, enabling the detection of security vulnerabilities at the pull request level even if the repository does not have CodeQL default settings enabled.

Topic: OpenAI Expands Everyday Support, Senior ChatGPT Workshops, and New Advertising Agent Launch

OpenAI is partnering with the American Association of Retired Persons (AARP) to offer free hands-on workshops in 10 U.S. cities, enabling 1,000 older adults to safely learn practical AI skills. At the same time, it is unveiling new tools for marketers and Sponsored Agents, along with new advertising features based on integrations with HubSpot and Shopify. I presented my experience.

Topic: Microsoft's Collaborative Efforts in Education and Trends in Research on AI Agent Consistency

Microsoft reaffirmed its commitment to providing technological support for educators, students, and education leaders to maximize learning outcomes, drawing on 50 years of experience in educational collaboration. Meanwhile, within the open-source ecosystem, IBM researchers shared their research on agent consistency evaluation via Hugging Face, among other contributions, as agents completed tasks with consistent performance. Methodologies for evaluating whether something can be reliably and consistently demonstrated are also being actively discussed.

Key Points to Watch

Topic: Level of Independence Assurance for Internal Safety Assessors and Transparency Verification

We must watch how the practical independence of safety evaluators deployed at Anthropic and the OpenAI Institute, as well as the scope of public disclosure of evaluation reports, will be determined. It is expected that whether external academia and regulatory bodies can secure a reliable level of information disclosure and objective oversight authority will become the key criterion determining the success or failure of future self-regulatory models.

Topic: Advancements in Next-Generation Accelerator Adoption and Dynamic Power Management Standards for Data Centers

With the market supply of NVIDIA's Vera Rubin NVL72, attention should be focused on how the dynamic power management technologies proposed by the AEMA alliance will be applied to large-scale AI data centers and power grid operations. It is necessary to observe what changes these efforts to overcome power supply constraints and maximize inference cost efficiency will bring to data center design standards.

Topic: Budget Control for Enterprise AI Tools and Deep Integration of Development and Training Operations

We need to examine how GitHub’s Copilot credit budget request flow support, AI scanning with reduced CodeQL dependencies, and Microsoft’s support for the educational ecosystem will impact actual user experience and cost control. It remains to be seen whether operational models that lower barriers to adopting AI tools in corporate and educational settings while enhancing management efficiency can take root. It draws attention.

Source

• TechCrunch AI: https://techcrunch.com/2026/09/16/anthropic-and-openai-want-to-embed-safety-evaluators-will-they-really-be-independent/
• TechCrunch AI: https://techcrunch.com/2026/09/16/anthropic-merges-claude-chat-and-cowork-in-one-interface/
• NVIDIA Blog: https://blogs.nvidia.com/blog/vera-rubin-nvl72-mlperf-inference/
• NVIDIA Blog: https://blogs.nvidia.com/blog/ai-energy-management-alliance/
• GitHub Blog: https://github.blog/changelog/2026-09-16-copilot-budget-increase-requests-are-generally-available
• GitHub Blog: https://github.blog/changelog/2026-09-16-code-scanning-ai-scan-no-longer-requires-codeql-default-setup
• OpenAI: https://openai.com/index/helping-older-adults-use-ai-in-everyday-life
• OpenAI: https://openai.com/index/reimagining-advertising-with-ai
• Microsoft Blog: https://blogs.microsoft.com/blog/2026/09/16/microsofts-commitment-for-ai-in-education/
• Hugging Face: https://huggingface.co/blog/ibm-research/altk-evolve-consistency