The Download: attempting to track AI, and the next generation of nuclear power

2026-02-05 · Source: MIT Technology Review · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Cybersecurity & Data Privacy, Emerging Technologies & Innovation · Depth: Fundamental Awareness, medium

Summary

The February 5, 2026 edition of The Download covers several key technology developments. A significant focus is on the "most misunderstood graph in AI" by METR, which tracks exponential growth in AI capabilities, noting that Anthropic's Claude Opus 4.5, released in November, significantly outperformed predictions by completing a five-hour human task independently. The brief also addresses next-generation nuclear power, providing answers to common questions from a recent online roundtable discussion. Additionally, it highlights Anthropic's new coding tools, which are impacting markets, and discusses Apple's Lockdown Mode effectively preventing FBI access to a reporter's iPhone. A critical concern is raised about a major open-source AI training dataset, DataComp CommonPool, which is estimated to contain hundreds of millions of images with personally identifiable information, including passports and credit cards, scraped from the web.

Key takeaway

For CTOs and VPs of Engineering evaluating AI model capabilities or data sourcing, understand that reported exponential growth in AI performance, like with Claude Opus 4.5, may mask underlying complexities. Simultaneously, your teams must exercise extreme caution regarding AI training data; the widespread presence of personally identifiable information in datasets like DataComp CommonPool necessitates rigorous data governance and privacy safeguards to mitigate significant legal and ethical risks.

Key insights

AI capability growth is often misunderstood, and large datasets contain vast amounts of scraped personal data.

Principles

• AI capabilities can exceed exponential growth trends.
• Online data is highly susceptible to scraping for AI training.

In practice

• Review AI model evaluation metrics critically.
• Assume all online personal data is publicly accessible.

Topics

• AI Capabilities Evaluation
• Large Language Models
• AI Training Data Privacy
• Nuclear Power
• AI Market Impact

Best for: CTO, VP of Engineering/Data, Director of AI/ML, General Interest, Tech Journalist, Business Analyst

Related on AIssential

• MLWhiz Weekly Recsys/ML/GenAI Newsletter # 6 — AI capabilities are rapidly outrunning existing institutional frameworks in cybersecurity, coding, and language design.
• The $200 Billion Data Debt: Every Major AI Lab Is Running Out of Fuel and Nobody Is Talking About… — The AI industry faces a critical data debt crisis as high-quality human-generated data is exhausted and synthetic data causes model collapse.
• Import AI 452: Scaling laws for cyberwar; rising tides of AI automation; and a puzzle over gDP forecasting — Advanced AI models are rapidly enhancing cyberattack capabilities and driving significant performance gains for startups.
• AI chatbots are giving out people’s real phone numbers — Generative AI chatbots frequently expose personal data, primarily due to PII in training sets, with limited user recourse.
• Weekly Review 26 December 2025 — AI's rapid expansion creates complex challenges across data, ethics, economics, and resource consumption.
• METR’s Joel Becker on exponential Time Horizon Evals, Threat Models, and the Limits of AI Productivity — META evaluates AI capabilities and propensities against threat models to assess societal risks and track continuous progress.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by MIT Technology Review.