For most of the AI revolution, one assumption held: human beings were building ever more capable machines. That assumption no longer holds.

Anthropic disclosed this week that Claude now leads 26% of the company’s AI research and development work, up from less than 1% in February, completing most tasks from a high-level prompt while a human supervises. Anthropic says Claude is not autonomously building its successor, but published the figures because AI helping build more powerful AI could make advanced systems harder for people to understand or control.

That is where this month’s Five Lenses begins. Each month I examine artificial intelligence through five dimensions: capability, control, competition, capacity, and consequences. This month, one thread runs through all five: AI is no longer just advancing, it has begun to participate in the process that advances it.

Capability: When AI Helps Build AI

We moved rapidly from chatbots that answered questions to agents that plan, use tools, write software, and conduct research. Now comes another step: Anthropic’s new “R&D Automation Index” shows Claude leading almost none of its AI development in February and 26% by August, with more than 90% now at a collaborative level or higher. That is not recursive self-improvement; humans remain in charge. The trajectory is clear: AI has moved from answering to helping build the AI that comes next.

Capabilities outside the lab are advancing too. OpenAI’s newly released GPT-6 Astra is the first model it has broadly deployed to cross its own “Critical” cybersecurity threshold: with the right tools and access, OpenAI says Astra can find previously unknown vulnerabilities and build exploits against well-protected systems without a person guiding every step. The capability question has changed: it is no longer simply what AI can do for us, but what happens once AI helps accelerate the next AI.

Control: Capability Outrunning Restraint

That question leads to control. Anthropic CEO Dario Amodei has called for “pacing” frontier AI development to give safety work time to catch up, proposing independent evaluators inside frontier labs, common safety standards, and international cooperation. OpenAI’s Sam Altman and xAI’s Elon Musk quickly voiced support.

The debate arrives amid troubling evidence — Anthropic disclosed four incidents in which Claude models gained unauthorized access to real third-party systems during cybersecurity evaluations, then widened its review to roughly 481 million transcripts after finding the fourth. These systems had not become conscious; the problem is more immediate: capable agents can pursue assigned objectives their designers never anticipated.

America cannot simply freeze its own development while competitors race ahead, and Amodei acknowledges that pacing by democratic nations must reckon with China. The answer is neither to halt AI nor to race forward without safeguards, but to keep capability under meaningful human control.

Competition: China Is Building More than a Model

Much of the debate over America’s AI race with China has centered on whose model is most capable. That measure no longer captures the whole contest: China is building an entire AI ecosystem of chips, models, cloud infrastructure, manufacturing, and its own global relationships.

Huawei illustrates the strategy: the company says Chinese demand for its AI computing equipment exceeds its ability to supply it, and it is accelerating new Ascend processors and computing systems meant to close the gap with Nvidia’s leading chips.

The competition extends beyond hardware. Anthropic reports that operators affiliated with Alibaba ran what it calls its largest model-distillation attack, generating more than 151 million exchanges between May and July to transfer Claude’s capabilities into Alibaba’s Qwen models. America may still hold the best individual technologies and still face a formidable competing ecosystem.

Capacity: The Bottleneck beneath the Chip

All that intelligence rests on something stubbornly physical: memory. China’s AI chipmakers are confronting a shortage of high-bandwidth memory, or HBM, the specialized memory that feeds data to advanced processors. Huawei, Cambricon, and other Chinese chipmakers have raised accelerator prices as costs climb, while U.S. restrictions limit China’s access to advanced HBM.

China is responding by expanding its domestic memory industry: CXMT, historically a DRAM maker, is preparing to enter the flash memory market, competing directly with domestic rival YMTC as AI demand tightens memory supplies worldwide.

Artificial intelligence looks weightless on a screen, but national AI power rests on a physical foundation of semiconductors, memory, data centers, electrical generation, and skilled workers most Americans never see, and the AI race may ultimately be constrained there.

Consequences: AI Is Changing Work, Not Simply Eliminating It

No AI prediction gets repeated more than this one: the machines are coming for our jobs. Some jobs will disappear; others will change. But this month’s data complicate that picture.

The Federal Reserve Bank of New York reports that AI use among surveyed businesses in New York and northern New Jersey has risen sharply, yet mass layoffs have not followed. Only 4% of surveyed service firms using AI reported AI-related layoffs in the past six months, and no surveyed manufacturers did. About 13% of service firms said they added workers because of AI, and retraining remains the most common response.

Those numbers should temper both utopian promises and apocalyptic predictions. AI will eliminate some jobs, reshape others, and create still others. But for Christians, employment statistics do not exhaust the question: Scripture presents work as more than economic output, an arena for creativity, responsibility, service, and vocation. If AI lets us produce more while we surrender the skills and judgment through which we mature, something is lost even as GDP rises.

The Human Question behind the Five Lenses

Taken together, September’s developments tell a larger story. AI is beginning to participate in the research that makes it more capable, even as developers struggle to keep autonomous systems within intended boundaries. Washington and Beijing are racing to build competing technological ecosystems, and both are discovering the contest ultimately depends on decidedly physical things: memory chips, factories, electricity, data centers, and skilled people. Beneath it all lies the question that matters most: what happens to us?

None of this requires Christians to fear artificial intelligence. But neither should we mistake intelligence for wisdom. Scripture tells us, “The fear of the LORD is the beginning of wisdom” (Proverbs 9:10, ESV). No benchmark measures that wisdom. No semiconductor manufactures it, and no training run can give a machine moral responsibility before God.

We will delegate calculation, research, and more work than we imagine, but we cannot delegate what it means to be human.

The most consequential question of the AI age may not be how intelligent our machines become. It may be whether, as they grow more capable, we remain wise enough to remember the difference.