OpenAI launches Multipath Reliable Connection as Nvidia ships Spectrum X

Key Takeaways

• OpenAI released Multipath Reliable Connection (MRC), an open-source networking protocol for large-scale AI training, developed with AMD, Broadcom, Intel, Microsoft, and Nvidia.
• MRC spreads data transfers across hundreds of paths, rerouting around congestion or failures in milliseconds, and is deployed across all of OpenAI's largest NVIDIA GB200 supercomputers.
• Nvidia's Spectrum X Ethernet fabric now supports MRC, and the protocol is available through the Open Compute Project.
• Inworld AI launched Realtime TTS 2, a voice model that hears full audio input, supports over 100 languages, and allows voice direction via plain language prompts.
• Coursera CEO reported that someone enrolls in an AI course every four seconds on the platform, which has over 100 million learners.
• Keeper Security research found 89% of IT leaders struggle with identity management, 43% cite AI machine identity management as a top gap, and 56% worry about employees exposing sensitive info to AI.
• Microsoft was named Overall Leader in KuppingerCole's 2026 Emerging AI Security Operations Center report for its Security Copilot and automatic attack disruption.
• In May 2025, law firm Latham and Watkins filed a court declaration with fabricated citations generated by Claude, leading to mandated AI disclosure and human verification in court filings.
• Nearly half of surveyed fleet managers use AI in their responsibilities; AI-powered routing can save at least 50% in fuel and operational costs.
• AI-driven freight automation could reduce a fleet's total cost of ownership by 8 to 13%, according to the 2026 State of Sustainable Fleets report.

OpenAI researchers say AI needs new supercomputer networks

OpenAI researchers Mark Handley and Greg Steinkrecker discussed the need for new supercomputer networks to handle AI training. They explained that traditional networking protocols struggle with the demands of training large AI models. The researchers highlighted their MRC system, which helps manage network congestion and hardware failures. MRC allows data to be rerouted around problems in milliseconds, making training more reliable and efficient.

OpenAI releases new training spec for large scale AI

OpenAI released Multipath Reliable Connection, an open source specification for large scale AI training networks. The spec was developed with AMD, Broadcom, Intel, Microsoft and Nvidia. It improves GPU performance and resilience in large training clusters by spreading data transfers across hundreds of paths. This allows data to be rerouted around congestion or failures in milliseconds. OpenAI has already used MRC to train multiple frontier models and made it available through the Open Compute Project.

OpenAI introduces MRC protocol for AI supercomputer networks

OpenAI introduced MRC, a new supercomputer networking protocol released through the Open Compute Project. MRC improves GPU networking performance and resilience in large training clusters by spreading a single transfer across hundreds of paths. It can route around failures in microseconds and runs simpler network control planes. MRC is already deployed across all of OpenAI's largest NVIDIA GB200 supercomputers used to train frontier models. The protocol has been used to train multiple OpenAI models with hardware from NVIDIA and Broadcom.

NVIDIA Spectrum X Ethernet now supports MRC protocol

NVIDIA announced that its Spectrum X Ethernet fabric now supports Multipath Reliable Connection, an RDMA transport protocol. MRC enables a single RDMA connection to distribute traffic across multiple network paths, improving throughput and load balancing for large scale AI training. OpenAI, Microsoft and Oracle are using this technology. MRC can detect network path failures and reroute traffic in microseconds, keeping thousands of GPUs synchronized during training. The protocol is now available as an open specification through the Open Compute Project.

Inworld AI launches Realtime TTS 2 voice model for conversation

Inworld AI launched Realtime TTS 2, a new voice model released as a research preview through its Inworld API. The model hears the full audio of the exchange, picking up the user's tone, pacing and emotional state. It operates as a closed loop system that takes actual audio as input, not just a transcript. The model includes four key features: voice direction using plain language prompts, conversational awareness, crosslingual support across over 100 languages, and advanced voice design from written prompts without reference audio.

Coursera CEO says someone enrolls in AI course every four seconds

Coursera CEO Gregory R. Floyd spoke at the HumanX conference about the rapid growth of AI education on the platform. He said that every four seconds, a new user enrolls in an AI course. Floyd highlighted the importance of upskilling and reskilling workers for technological change. He noted that Coursera has over 100 million learners on its platform. The CEO expressed optimism about AI's potential to drive economic growth while acknowledging the need for responsible AI development.

UNCTAD to examine AI and geopolitical shifts in global investment

The UN Conference on Trade and Development will hold a session to assess what AI driven investment shifts mean for developing economies. The session will examine how cross border investment decisions are increasingly shaped by supply chain resilience, technological security and trade barriers. The concept note links fast growing investment in AI and digital infrastructure to industrial policy priorities and national security concerns. Member states are invited to submit expert papers before the session, which will be held in person with a live audio stream.

Keeper Security research finds 89 percent of IT leaders struggle with identity management

Keeper Security released research showing that 89 percent of senior IT leaders find managing the growing identity footprint challenging. The study surveyed 3,200 cybersecurity decision makers across the United States, Europe, Asia Pacific and the Middle East. It found that 96 percent of organizations have disconnected or poorly integrated security tools creating exploitable gaps. Additionally, 43 percent of respondents identified AI related machine identity management as a top governance gap. The research also showed that 56 percent of respondents are concerned about employees exposing sensitive information to AI systems.

Microsoft named overall leader in AI Security Operations Center report

Microsoft was named an Overall Leader and Market Leader in KuppingerCole Analyst's 2026 Emerging AI Security Operations Center report. The report highlights how security operations are entering a new phase where automation is a core component of cybersecurity. Microsoft is investing in agentic security operations with innovations like Security Copilot and automatic attack disruption. These agents are designed to work alongside analysts, helping with reasoning across identity, endpoint, cloud and network signals. The goal is to help SOC teams operate with greater speed, confidence and resilience.

MCP Apps introduces new standard for AI chat interfaces

Liad Yosef and Ido Salomon of MCP Apps detailed their vision for a new standard for building interactive UI components within AI chat interfaces. MCP Apps enable applications to send their own UI elements directly to AI agents, allowing users to interact with rich dynamic components within their chat interface. The standard is designed to work across platforms and is agnostic to the UI generation method. Major companies like Microsoft, Google and GitHub are already integrating or supporting the standard. The goal is to create a more unified and interactive digital experience where applications can be seamlessly integrated into AI interfaces.

Claude hallucination in court raises questions about attorney liability

In May 2025, law firm Latham and Watkins filed a court declaration in a case against Anthropic that contained fabricated citation details generated by Claude, the AI model they were defending. The lawyer used Claude to format a legal citation, but Claude got the authors and title wrong. The error was not caught until opposing counsel investigated. The court mandated explicit disclosure of AI usage and human verification requirements for future filings. The incident raises questions about attorney responsibility under Rule 11 of the Federal Rules of Civil Procedure when using AI tools.

BEYOND HACK DAY opens applications for AI Innovative Hardware track

BEYOND HACK DAY has officially opened applications for its AI Innovative Hardware track at BEYOND Expo 2026. The challenge is built around the theme Empowering the Physical World with AI Agents and calls on global developers and creators to build working prototypes. The event is structured as a 72 hour intensive creation cycle from May 28 to May 30. Projects will be evaluated across five dimensions including innovation, practicality and technical completion. Award winning teams will receive a fast track entry into the JLC Technology Group's Starfire Innovation Competition with million dollar prizes.

Trucking fleets use AI for route optimization and cost savings

Industry leaders from top tech and transportation companies gathered at ACT Expo in Las Vegas to discuss AI in the trucking industry. Nearly half of surveyed fleet managers report using AI in their responsibilities as of late 2025. AI powered routing can save at least 50 percent in fuel and operational costs by adapting routes in real time based on weather, traffic and construction history. Other uses include predictive maintenance, safety systems and freight operations. The 2026 State of Sustainable Fleets report found that AI driven freight automation could reduce a fleet's total cost of ownership by 8 to 13 percent.

Sources

• OpenAI Podcast: AI Needs New Supercomputer Networks
• OpenAI Launches Training Spec to Boost Large-Scale AI
• Unlocking large scale AI training networks with MRC (Multipath Reliable Connection)
• NVIDIA Spectrum-X — the Open, AI-Native Ethernet Fabric — Sets the Standard for Gigascale AI, Now With MRC
• Inworld AI Launches Realtime TTS-2: A Closed-Loop Voice Model That Adapts to How You Actually Talk
• Coursera CEO - enrollment in AI course ‘every four seconds’
• UNCTAD to examine AI and geopolitical shifts in global investment
• Keeper Security Research Reveals 89% of IT Leaders Struggle to Manage Growing Identity Footprint Amid AI Expansion
• ​​Microsoft named an overall leader in KuppingerCole Analyst’s 2026 Emerging AI Security Operations Center (SOC) report ​​
• MCP Apps: Building the Future of AI Interactions
• When Claude Hallucinates in Court: The Latham & Watkins Incident and What It Means for Attorney Liability
• BEYOND HACK DAY Official Announcement: Call for AI Innovative Hardware
• How are trucking fleets using AI?