The Artificial Intelligence (AI) sector has entered a phase of unprecedented acceleration, with the line between innovation and risk blurring like never before.
Anthropic, until now seen as the “voice of conscience” of Silicon Valley, this week took a step that signals a paradigm shift: updating its Responsible Scaling Policy (RSP 3.0). By abandoning the commitment to pause model development if security risks were high — unless its competitors do the same. The company admits that commercial survival is taking precedence over caution.
According to Anthropic’s official document (RSP 3.0), released on Tuesday (February 24), the North American company justifies this change by pointing to a “strategic environment where technological leadership is inseparable from global security.” In practice, this means that the company led by Dario Amodei will not accept being left behind while others advance without the same ethical constraints.
“Visibility void” in metrics
This decision comes at a time when the industry itself admits that AI is developing at a speed that traditional tools can no longer keep up with. According to an analysis published by The Telegraphprogress is so rapid that “We’re running out of ways to measure what these machines can do.”
“Frontier” models are achieving near-perfect results in benchmark tests (benchmarks) like the MMLU, which experts predicted would take years to overcome.
This “metrics crisis” creates a visibility void. As a report from the AI Safety Institute also this week notes, “the gap between actual capabilities and our ability to test them is growing dangerously.” Without reliable metrics, it becomes nearly impossible to assess the risks of systems that now use “complex reasoning” and whose internal processes remain opaque to their own creators.
Defense Pressure
Geopolitical pressure is also shaping this new scenario. The US Department of Defense has been vocal in the need to integrate these tools, warning that “excessive prudence could constitute a national security risk in the face of the advance of rival powers”, according to a Pentagon statement released this Wednesday by international media.
It is in this context that Anthropic, under pressure from investors, decided that its security mission must now be balanced with “competitive viability in the global market”.
Europe in the regulatory maze
On this side of the Atlantic, the European Union is trying to maintain a difficult balance with the EU AI Act. For many critics, these standards may well represent “economic suicide”, but the European Commission takes a different view, with Eurocrats and other policymakers reiterating that the aim is not to halt innovation, but rather to ensure that Europe is “the world leader in trusted AI”.
Leave a Reply