Anthropic says AI systems moving towards building themselves - The Daily Star

TEXT START: Artificial intelligence giant Anthropic said yesterday that AI systems are increasingly capable of building future versions of themselves, adding to mounting concerns about the dangers of the powerful technology.

THE DISSECTION

This is a corporate warning disguised as public transparency. Anthropic is disclosing that Claude already performs 26 percent of its research and development work, collaborates on more than 90 percent of tasks, and operates alongside roughly 30,000 agents. The article frames the central issue as whether AI will become fully autonomous and uncontrollable. The more immediate reality is uglier: AI is already entering the machinery that produces AI, reducing the human labor required to reproduce the system.

The IPO detail exposes the contradiction. Anthropic is warning about acceleration while preparing to monetize it. The warning functions as risk disclosure, reputation management, and pre-emptive insulation from responsibility.

THE CORE FALLACY

The text treats recursive self-improvement—a model autonomously building its successor—as the decisive threshold. Under the Discontinuity Thesis, that threshold is unnecessary. The economic break begins when AI becomes cheaper and more capable than humans across enough cognitive work to sever the labor-to-wage-to-consumption circuit.

Claude does not need to be fully autonomous to make researchers, engineers, and other cognitive workers economically redundant. Human supervision can be a temporary interface, not durable human control. The article mistakes the final dramatic phase of automation for the beginning of the crisis.

HIDDEN ASSUMPTIONS

• Human supervision will remain substantive rather than becoming nominal approval over machine-generated work.
• Governments and third parties can slow development despite competitive pressure and concentrated AI ownership.
• Public visibility produces control. It does not; disclosure without enforcement is an autopsy report published while the patient is still being harvested.
• The primary danger is loss of control over powerful systems, rather than loss of productive participation by the majority.
• Regulation can manage the transition without confronting who owns the models, compute, energy, and resulting output.
• AI-driven R&D gains will remain confined to AI companies instead of diffusing through the broader economy.

SOCIAL FUNCTION

Classification: transition management, elite self-exoneration, and partial truth.

The article admits enough of the mechanism to establish credibility—AI is already performing core development work—then shifts attention toward spectacular autonomy, environmental costs, and hypothetical containment failures. That framing turns an ownership and labor crisis into a safety-management problem. Anthropic appears as the responsible observer of a process it is actively accelerating and commercializing.

It is not pure propaganda. The reported metrics are materially significant. But the narrative sterilizes their economic consequence. It presents the machine’s growing capacity to reproduce itself as a future hazard instead of recognizing it as present evidence that human cognitive labor is being displaced from the production chain.

THE VERDICT

This is an early-warning memo from inside the replacement machine. It provides evidence of cognitive automation advancing into high-value R&D, but it overfocuses on recursive self-improvement and underreports the immediate structural threat: the progressive removal of humans from economically necessary work. The article does not prove total system collapse, but under the DT lens it shows the machinery moving toward P1 while institutions perform concern as a substitute for coordination. The countdown is not waiting for an autonomous successor. It has already begun with supervised systems that make fewer humans necessary.