Anthropic Withdraws Security Promise: A Major Turning Point in the AI World

Anthropic, known in the AI industry as a pioneer in ethical and safe development, has made one of the most significant decisions in its history: it has officially rescinded its flagship safety commitment, the Responsible Scaling Policy (RSP). This policy, introduced in 2023 and designed to control safety risks during the scaling of AI systems, had served as a global benchmark among academic, industrial, and regulatory circles. Now, however, the link to this policy has been removed from the company’s official website. The removal of the RSP is not merely the deletion of a webpage; it signals a fundamental philosophical and strategic shift in the very foundation of AI safety understanding.

Why This Decision? Just Market Pressure?

Anthropic’s move appears puzzling at first glance. When launching Claude 2 in 2023, the company had earned trust from regulators in the U.S. and EU by assuming “transgressive responsibility” regarding “malicious use” and “multi-system risks.” The RSP mandated pre-approval for developing models with computational costs exceeding $10 billion—a clear differentiator from competitors like Google, OpenAI, and Meta. So why is such a strong brand abandoning such a robust ethical pillar?

Analyses based on internal sources reveal three primary reasons behind this decision. First, speed of production and competitive pressure. Models like OpenAI’s GPT-4o and Google’s Gemini Ultra have been rapidly deployed to market. Anthropic began to believe that the RSP prolonged development timelines and delayed meeting customer demands. Second, regulatory uncertainty. While federal regulations in the U.S. remain unclear, the EU’s AI Act imposes strict limits, particularly on high-risk systems. Anthropic has stated that certain provisions of the RSP conflict with EU legal requirements, thereby creating legal risk. Third, and most critically, an internal shift in belief: Some of the company’s senior researchers have begun arguing that the concept of “supervised scaling” actually hinders AI progress and that there is no inherent trade-off between “development speed” and “safety.”

What Does This Mean? Not Just Anthropic, But the Entire Industry

Anthropic’s decision may go down in history as an “ethical breaking point” in the AI world. Because the RSP was not merely a corporate policy—it was the most concrete embodiment of the “safe AI” philosophy. The policy required developers to undergo third-party security audits before releasing models, test for “multi-hazard scenarios,” and publish transparent reports. Now, by abandoning this framework, Anthropic has signaled its abandonment of the “self-regulating” model in favor of a strategy driven solely by “market and legal mandates.”

This is a major signal to ethics-focused regulators in the EU and Canada. Anthropic was previously the most trusted partner in recent years. Now, it demonstrates that a company can treat its ethical pledge as merely a “marketing tool” under competitive pressure. This could set a precedent for other companies to follow suit. OpenAI’s recent relaxation of its security protocols for its 2024 “superuser” models may already mark the beginning of this trend.

What Are Users Losing?

Ordinary users will not directly feel the impact of this change. Claude 3.5 will still behave like a “friendly” and “safe” assistant. But behind the scenes, the model’s boundaries have become more flexible. For example, under the RSP, a model would have rejected requests such as “generate destructive code” or “extract sensitive data.” Now, these filters may be less stringent. Developers, meanwhile, will gain more freedom—but also greater responsibility. With this change, Anthropic has begun defining “safety” as a “user responsibility.” In other words: “We won’t protect you—you protect yourself.”

What’s Ahead?

Anthropic’s move could trigger a wave of “ethical deregulation” across the AI industry. But it will also compel governments to enact stricter laws. The EU may classify this development as a “public safety threat” and suspend Anthropic’s services in Europe. In the U.S., this decision could accelerate the drafting of AI Safety Acts in Congress. Company leaders describe this decision as a “dynamic strategy.” But the truth is: when a company abandons its ethical pledge, it doesn’t just lose a policy—it loses trust.

Anthropic was born to make technology “safe.” Now, it appears to have been reborn to make technology “fast.” This transformation questions not just one company’s decision, but the very foundation of AI’s relationship with humanity.