One of the most dramatic shifts! Anthropic abandons its signature "safety guardrail" commitment

Anthropic has long positioned itself as a safer alternative to other artificial intelligence competitors, but now it is relaxing its commitment to maintaining safety "guardrails." This is one of the most notable policy shifts in the AI industry so far—a startup once driven by the mission to "benefit humanity" is now turning its focus to profitability and success.

Anthropic stated in its 2023 "Responsible Scaling Policy" that it would delay relevant research and development if AI advancement might pose dangers. However, in a blog post published this Tuesday, Anthropic said it was updating its rules: if it believes it does not maintain a significant technical lead over competitors, it will no longer delay development for this reason. Anthropic’s statement says:

“The policy environment has shifted to prioritize AI competitiveness and economic growth, while safety-focused discussions have yet to achieve substantive progress at the federal level.”

An Anthropic spokesperson said: “From the beginning we've stated that the speed of AI development and the uncertainty of the field will require us to quickly iterate and improve our policies.”

Analysis points out that Anthropic’s move highlights the increasingly frequent clash between the lofty ideals guiding early AI startups and mounting pressures for profitability and competition. Anthropic is competing for dominance in this revolutionary technology with strong rivals, including OpenAI, Google, and Musk’s xAI company.

Anthropic CEO Dario Amodei once worked at OpenAI, and he left in 2020, partly because he was concerned that the company was prioritizing commercialization and development speed over safety.

OpenAI was originally a nonprofit organization, and last year transitioned to a more traditional for-profit enterprise. In 2024, it also updated its mission statement, removing the word “safely,” whereas its previous goal was “to ensure that artificial general intelligence safely benefits humanity.”

Both OpenAI and Anthropic are now pushing to go public as early as this year, hoping to capture investors’ interest in AI. Anthropic’s recent valuation is $380 billion, and OpenAI is fundraising at a valuation of over $850 billion.

This policy update from Anthropic comes amid escalating disputes with the U.S. Department of Defense. The controversy centers on Anthropic’s insistence on setting safety guardrails for its Claude AI tool. On Tuesday, the Pentagon threatened that unless Anthropic accepted government terms by Friday, it would invoke a Cold War-era law to force the company to allow the military to use its technology.

According to media reports, during talks on Tuesday between Amodei and U.S. Secretary of Defense Pete Hegseth, U.S. officials laid out several potential consequences, including threatening to classify Anthropic as a supply-chain risk and invoking the Defense Production Act to forcibly use its AI software, even without the company’s consent.

Earlier this month, Anthropic senior safety researcher Mrinank Sharma announced his departure. In a letter posted to the X platform addressed to colleagues, he wrote: “I keep reflecting on the situation we’re in. The world is in danger. And it’s not just from AI or biological weapons, but from a range of interconnected crises happening at this moment.”

It’s not just Anthropic and OpenAI struggling with AI safety measures. Earlier this month, media reported that Musk’s SpaceX and its subsidiary xAI are involved in a secret new Pentagon bid project, developing voice-controlled autonomous drone swarm technology. This is seen as a controversial shift. Previously, Musk opposed making “new killing machines.” He even sued OpenAI, expecting it to remain nonprofit and committed to building safe AI for the public interest.

According to previous reports, OpenAI also supports Applied Intuition’s proposal in the drone bidding project, but its participation will be limited to the “mission control” aspect—that is, converting battlefield commanders’ voice and other instructions into digital commands.

The conflict between Amodei and OpenAI CEO Sam Altman has also sometimes gone public. At an AI summit in New Delhi last week, while both stood beside Indian Prime Minister Modi, others on stage held hands in a line, but the two refused to join hands.

Risk Warning and DisclaimerThe market carries risks and investment must be made with caution. This article does not constitute personal investment advice and does not take into account individual users’ specific investment goals, financial situations, or needs. Users should consider whether any opinions, viewpoints, or conclusions in this article are suitable for their particular circumstances. You invest at your own risk based on this.