The accelerated advancement of intelligent systems could cause humanity to lose control over technology. The alert was released this Thursday (4) by The Anthropic Institute in a self-improvement report.
The company revealed that the artificial intelligence already writes more than 80% of the code integrated into its own system. The unprecedented internal data raised the alarm about the speed of the machines’ autonomy.
Currently, the company’s engineers deliver, on average, eight times more code per quarter than they recorded before. This acceleration drastically reduces dependence on strictly human labor.
Risk of total and recursive autonomy
The technical trend points to the so-called recursive self-improvement. The phenomenon occurs when artificial intelligence becomes capable of fully autonomously designing and training its own technological successor.
According to Anthropic’s official document, this autonomy may come faster than institutions are prepared. Without direct human oversight, traditional security and monitoring methods lose their effectiveness.
The pace of execution of complex tasks completed without human intervention now doubles every four months. In global software engineering exams, systems reached complete saturation in just two years.
Superhuman ability in testing
The tool Claude Mythos Preview achieved an optimization capacity considered superhuman. The system achieved a 52-fold acceleration in internal tests for training small artificial intelligence models.
In complex missions without detailed specifications, the software’s success rate reached 76% in May 2026. The index represents a significant increase of 50 percentage points in six months.
Security review of new codes has also been delegated to an intelligent evaluator. The automated tool was able to intercept a third of technical errors in retrospective incident analyses.
Global truce proposal
Anthropic suggests that governments and cutting-edge laboratories create a global joint verification mechanism. The structure would allow monitoring whether competing companies are complying with strict security standards.
The report proposes applying supervised temporary pauses in global development if risks spiral out of control. The holding claims that a unilateral pause would only change the leader of the technology race.
The company plans to organize debates in the coming months with public policy makers to discuss systems self-improvement. The objective is to involve civil society in this urgent deliberation.
Source: www.olhardigital.com.br
Source link
