Anthropic Uses AI to Build Better AI Models, Raising Safety Questions

Exercise 1
Vocabulary
Repeat each word, its meaning and the example sentence out loud.
frontier AI
noun phrase/ˈfrʌntɪər eɪ aɪ/the most advanced artificial intelligence research and systems currently being developed, representing the cutting edge of the field
Our research division focuses on frontier AI applications, which means we need to stay abreast of both capabilities and ethical implications.
pump the brakes
phrasal verb/pʌmp ðə breɪks/to slow down or reduce the pace of something; to apply caution to a process that is moving too quickly
The CEO has been calling on the industry to pump the brakes on deploying unvetted models, but our competitors show no signs of listening.
redact
verb/rɪˈdækt/to remove or obscure sensitive information from a document before publishing it publicly
The legal team redacted proprietary pricing information from the technical report before we shared it with external stakeholders.
diminishing returns
noun phrase/dɪˈmɪnɪʃɪŋ rɪˈtɜrnz/a situation where additional effort produces progressively smaller improvements or benefits, eventually making further investment uneconomical
We're beginning to see diminishing returns on our current infrastructure; each upgrade yields less performance gain than the last.
commerciallycompetitive sensitivity
adjective phrase/kəˈmɜrʃəli ˈsɛnsɪtɪv/information that could give rivals an unfair advantage if disclosed, making it strategically sensitive to a company's market position
We cannot publish our benchmarking methodology because it contains commercially sensitive data that competitors could exploit.
Exercise 2
Article
Read the article aloud on your own, or listen to it first.
Anthropic Uses AI to Build Better AI Models, Raising Safety Questions
Tap any word to see what it means.
A safety research organization has published findings showing that Anthropic is deploying artificial intelligence systems to accelerate its own AI development.
According to a report from METR following an 11-week investigation inside Anthropic, the company's AI models are improving at roughly twice the rate they were a year ago. The researchers found that this acceleration stems primarily from AI systems assisting human researchers in their work, rather than simply scaling up computing power—the traditional driver of AI progress.
Anthropic employees report extraordinary productivity gains when using the company's most sophisticated internal models. Staff members indicated they are approximately 12 times more productive with AI assistance than they would be working independently, a substantial jump from roughly six times the productivity reported just months earlier with previous systems.
What makes this finding particularly striking is that much of this acceleration comes from models the company has not released publicly. Over the past six months alone, Anthropic completed training on three internal models beyond what the general public can access.
The timing creates a notable paradox. Anthropic's CEO has been vocal about calling on the AI industry to pump the brakes on development.
Yet internally, the company is dramatically accelerating its research velocity precisely through AI-driven automation.
METR researchers found no clear signs of diminishing returns or observable obstacles that might slow this acceleration. They also warned that autonomous AI contributions—where systems develop themselves with minimal human oversight—appear poised to play an increasingly significant role.
However, the report's credibility faces a substantial caveat: Anthropic redacted a majority of the evidence METR collected before releasing the findings publicly. The company cited both competitive sensitivity and safety concerns about sharing technical details that could inform rivals.
METR emphasized that these redactions severely compromise the public's ability to independently evaluate the organization's conclusions. The researchers noted they had submitted roughly 100 pages of detailed findings, but what the public received represented only a fraction of that evidence.
Anthropic's cofounder acknowledged the redactions were extensive, arguing the company made the correct choice by withholding information that could accelerate competitors' progress. He maintained that transparency in this domain carries genuine risks.
The situation raises uncomfortable questions about how much accountability and oversight can genuinely function when companies control what information gets disclosed about their own development processes. As Anthropic approaches its public offering, this tension between transparency and commercial protection becomes increasingly consequential for investors and regulators alike.
Discuss this story out loud with our AI speaking coach.
It asks one question at a time, at your level, and gives you feedback at the end.
Start the discussion Retell it3,000+ learners
Also works inChatGPTandClaude
Exercise 3
Discussion
Have a discussion based on the following questions.
- 1.In your own organization, how do you balance the pressure to move quickly with the need to maintain oversight and safety in AI development?
- 2.Do you think companies should be obligated to disclose detailed information about their AI acceleration methods, or is competitive confidentiality justified in this context?
- 3.If Anthropic's internal models are substantially more capable than their public releases, what implications does this have for fair competition in the AI industry?
- 4.How should regulators approach situations where companies are simultaneously advocating for industry-wide caution while internally accelerating their own development?
Exercise 4
Further discussion
Have a discussion based on the following questions.
- 1.What mechanisms could enable genuine independent oversight of frontier AI development without compromising legitimate competitive interests?
- 2.As AI systems become increasingly capable of improving themselves, how should responsibility for safety outcomes be distributed between researchers, companies, and regulators?
- 3.Is the current disclosure model for AI research adequate, or do we need new frameworks that balance transparency with commercial protection?
Discuss this story out loud with our AI speaking coach.
It asks one question at a time, at your level, and gives you feedback at the end.
Start the discussion Retell it3,000+ learners
Also works inChatGPTandClaude
Source
This lesson is our own retelling, for English learners, of a story from The Verge.
Related stories

A Mathematician and an AI System Both Claim to Have Solved the Riemann Hypothesis
22 November 2026 · C2

Anthropic's IPO: Can a safety-focused AI company survive shareholder pressure?
22 November 2026 · C1

Self-Teaching AI Systems: A Breakthrough or a Risk?
22 November 2026 · B2

AI and sensors help warn communities about dangerous landslides
22 November 2026 · B1



