Amazon launches G7e instances as Honor robot Blitz wins half marathon

Key Takeaways

Chinese Robot Beats Human Record in Beijing Half-Marathon

A humanoid robot developed by Honor won the second annual robot half-marathon in Beijing on April 19, 2026. The robot finished the 21-kilometer race in 50 minutes and 26 seconds, which is seven minutes faster than the human world record. This performance marked a huge improvement from last year when the winning robot took over two hours. The race took place in Beijing E-Town alongside 12,000 human runners on separate tracks. About 40% of the robots navigated the course using autonomous navigation while others were remotely controlled.

Honor Robot Surpasses Human Record in Beijing Race

On April 19, 2026, a humanoid robot from Honor set a new half-marathon record in Beijing by finishing in 50 minutes and 26 seconds. This time beat the human world record held by Jacob Kiplimo from Uganda, who finished in about 57 minutes. The event featured over 100 robots from 76 institutions competing against 12,000 human runners. The winning robot, named Blitz, had long legs of about 95 centimeters and used a liquid-cooling system to prevent overheating. Du Xiaodi, a test development engineer for Honor, noted that the design was modeled after elite human athletes.

Robot Runner Breaks Half-Marathon Record in China

Humanoid robots from Chinese teams outran human competitors during a half-marathon event in Beijing on April 19, 2026. The fastest robot from Honor completed the 13-mile course in 50 minutes and 26 seconds while operating autonomously. The race included 300 robotic contestants from about 100 teams running alongside 12,000 human runners on parallel tracks. The winning robot used long legs measuring approximately 37 inches and incorporated advanced balance systems. Although some robots fell or needed assistance, the event showed rapid progress in robotic speed and autonomy.

Lightning Robot Wins Half-Marathon in Under an Hour

An autonomous robot named Lightning finished a 13-mile half-marathon in Beijing in 50 minutes and 26 seconds on April 19, 2026. This time was less than half the duration of last year's fastest robot, which took two hours and 40 minutes. Lightning beat all 12,000 human runners and the fastest human runner this year, who finished in one hour and seven minutes. The robot operated without direct human control and beat the previous record by a significant margin. The event demonstrated that machines can outperform humans at specific physical tasks under controlled conditions.

Amazon Launches New G7e Instances for AI Inference

Amazon SageMaker AI announced the availability of G7e instances powered by NVIDIA RTX PRO 6000 Blackwell GPUs on April 20, 2026. These instances offer 96 GB of GDDR7 memory per GPU and provide up to 2.3x inference performance compared to the previous G6e generation. The new hardware supports large language models with up to 300 billion parameters on an 8-GPU node. Key features include 1,600 Gbps networking throughput and 768 GB of total GPU memory on the largest instance size.

NVIDIA Uses FP8 Precision for Faster AI Training

NVIDIA researchers developed a method to run reinforcement learning training with end-to-end FP8 precision to improve performance. This approach uses low-precision data types to boost throughput and reduce memory usage during the generation and training phases. The new recipe reduces numerical disagreement between the generation and training engines compared to using only FP8 for generation. Experiments show that this method completely closes the accuracy gap found in previous approaches using mixed precision.

Most CISOs Lack Visibility Into AI Usage

A report by Pentera found that zero CISOs have full visibility into how AI operates across their organizations. Sixty-six percent of CISOs report limited visibility into AI usage, and 75% rely on traditional security tools that were not designed for AI ecosystems. Web-facing assets are the most common entry point for breaches, with attackers often moving from web assets to AI systems. The main barriers to securing AI are a lack of internal expertise and limited visibility rather than budget constraints.

Organizations Spend Hours Securing AI-Generated Code

A Cloudsmith report found that 58% of organizations spend over 10 hours per month validating and securing AI-generated code. Thirty-one percent of organizations using AI-generated code spend 10 hours or less per month on this task. Concerns include risks like slopsquatting where models hallucinate non-existent package names, and hidden vulnerabilities in software supply chains. Many organizations struggle to provide provenance data for audits within the required 48-hour timeframe after a breach.

UK Courts Discuss AI Use in Legal Disclosure

Legal representatives in England and Wales must use technology to promote efficient disclosure under Practice Direction 57AD, but generative AI is not specifically mentioned in the current rules. Generative AI can help with document classification, privilege review, and redactions in legal disclosure processes. However, challenges include ensuring cooperation between parties and defining the scope of disclosure regarding prompts used with AI. The courts are considering whether further guidance is needed to address the rapid development of new technologies.

Healthcare AI Safety Gaps Go Beyond Cybersecurity

Dr. Seixas argues that AI safety in healthcare involves more than just protecting systems from cyberattacks. He introduces the PAST framework which covers poisoning, abuse, stealing, and tricking, but emphasizes a second form of safety focused on human impact. Three major safety gaps exist: model drift as the world changes, misuse of AI in settings it was not built for, and opacity where models cannot explain their decisions. Healthcare leaders must focus on these issues throughout the life cycle of AI tools.

AI Predicts NFL Draft Picks After Lawrence Trade

AI generated mock drafts for the 2026 NFL Draft following the Dexter Lawrence trade, with the Las Vegas Raiders selecting first. The AI picked Fernando Mendoza as the quarterback for the Raiders and David Bailey as an edge rusher for the New York Jets. Other top picks included Arvell Reese for the Arizona Cardinals and Jeremiyah Love for the Tennessee Titans. The predictions were based on player talent and team needs after the major trade occurred.

DeviQA Creates New Practice for AI Testing

DeviQA launched a dedicated AI and machine learning testing practice on April 20, 2026, to address quality issues in non-deterministic systems. The company developed the AIVT Framework which focuses on output consistency, behavioral boundaries, safety, and bias exposure. This methodology is designed for product teams building AI-powered applications like recommendation engines and intelligent automation workflows. Traditional QA scripts are insufficient for systems where behavior changes with model updates and context.

Musk Proposes Universal Income to Prevent Deflation

Elon Musk argued that issuing dollars to people is necessary to prevent massive deflation if AI and robots increase output significantly. He suggested that federal funding for universal high income checks is the preferred response to job losses caused by automation. Economist Sanjeev Sanyal criticized the plan as economically unsound and risky for government finances. The debate highlights the broader implications of automation on the economy and the need for new economic policies.

AI Tools Enable New Forms of Workplace Discrimination

Artificial intelligence has enabled new methods of discrimination in employment law, particularly through deepfakes and automated harassment bots. These tools can create realistic content that impersonates coworkers or executives to harass targets. Automated interview tools also analyze facial expressions and tone, often misinterpreting cultural differences as low confidence. This leads to a lack of diversity in the workforce and creates new obstacles for victims seeking justice.

Sources

• China Robot’s 50-Minute Half Marathon Shows Pace of AI Progress
• A humanoid robot sprints past the human half-marathon world record in Beijing race
• Robot runner handily beats humans in half-marathon, setting new record
• Humanoid ‘Lightning’ robot smashes the half-marathon record
• Accelerate Generative AI Inference on Amazon SageMaker AI with G7e Instances
• Run High-Throughput Reinforcement Learning Training with End-to-End FP8 Precision
• The AI Visibility Gap Is Real – And It Lives on Your Website
• 58% of Organizations Spend Over 10 Hours a Month Securing AI-generated Code
• Disclosure in England and Wales: Where to Next on the AI Journey?
• Beyond Cybersecurity: The Biggest AI Safety Gaps in Clinical Care
• 2026 NFL mock draft: AI predicts 1st round after Dexter Lawrence trade
• DeviQA Launches Dedicated AI/ML Testing Practice to Address Growing Demand for Reliable AI-Powered Products
• Elon Musk Says 'If AI/Robots Increase Output, Then You Must Issue Dollars To People or There Will Be Massive Disinflation'
• Fighting Discrimination In The Age Of AI