This Week in AI: Security Flaw Exposes AI Giants While Robot Workers Get Upgrade

This week in AI, security researchers stole artificial intelligence models with near-perfect accuracy by capturing electromagnetic signals, as MIT unveiled a robot system that can handle odd-shaped warehouse packages with 80% success.

Meanwhile, three nations launched AI regulations, and Google released its more autonomous Gemini 2.0. Tech leaders acknowledge tougher challenges ahead as some AI models make dramatic leaps while others hit roadblocks.

AI Models Vulnerable to Electromagnetic Signal Theft

North Carolina State University researchers demonstrated a security vulnerability in AI systems, achieving over 99% accuracy in extracting AI models by capturing electromagnetic signals from computer hardware.

The technique, which doesn’t require direct system access, raises alarms for tech giants like OpenAI, Anthropic and Google, which have invested heavily in proprietary AI models. The discovery highlights growing cybersecurity challenges as businesses increasingly rely on AI for competitive advantage.

MIT Develops AI System for More Capable Warehouse Robots

MIT researchers created PRoC3S, a new AI system that could help warehouse robots handle odd-shaped packages and navigate crowded spaces more effectively.

The system combines AI language models with computer vision and tests actions in a virtual environment before executing them. In lab tests, it completed basic tasks like drawing shapes and sorting blocks with 80% accuracy.

The technology aims to help robots perform complex warehouse jobs that typically require human dexterity.

Three Nations Advance AI Oversight With Different Approaches

A bipartisan U.S. House task force recommended industry-specific AI oversight rather than broad federal regulations, marking Congress’s first comprehensive framework.

Meanwhile, Malaysia established a National AI Office to coordinate policy and development as it positions itself as a tech hub.

And, the United Kingdom introduced a consultation on copyright reforms to balance AI innovation with creative industry protections.

The moves reflect growing global efforts to establish AI governance frameworks.

Google’s Gemini 2.0 Promises More Autonomous AI Operations

Google released Gemini 2.0, an AI system designed to handle complex tasks across multiple platforms with greater autonomy.

The system powers projects like Astra for Android devices and Mariner for web navigation, suggesting a shift from command-based AI to more independent operation.

A key feature is its unified approach to processing different types of information, integrating text, images and audio handling that previously required separate tools.

AI Progress Shows Mixed Signals as Tech Giants Navigate Future

AI is delivering contrasting signals, with some models achieving dramatic leaps while others hit unexpected roadblocks.

Tech leaders acknowledge that while AI isn’t stalling, future progress faces steeper challenges. As companies race to develop more practical AI tools, businesses are carefully weighing investment decisions against a technology curve that defies simple characterization as either accelerating or slowing.

For all PYMNTS AI coverage, subscribe to the daily AI Newsletter.

The post This Week in AI: Security Flaw Exposes AI Giants While Robot Workers Get Upgrade appeared first on PYMNTS.com.