Seizing power in the name of "safety"? Unpacking Anthropic's contradictory mindset: Only when we dominate will the world be safe.

According to an in-depth feature article published by WIRED on June 25, 2026, Anthropic, an AI giant founded by former OpenAI employees, is displaying a highly contradictory corporate identity. On one hand, they loudly warn of the large-scale destruction risks AI could bring; on the other, they actively pursue a massive market value of nearly $1 trillion, even securing military contracts with the U.S. Pentagon. Facing external criticism, Anthropic internally adheres to a core philosophy: "Only if we achieve absolute success can we ensure AI safety."
(Previous context: Anthropic accuses Alibaba of launching the "biggest clone attack in history," spamming Claude 28.8 million times)
(Background supplement: Former Anthropic researcher founds Mirendil, raising $200 million, claiming to create "self-upgrading AI")

Table of Contents

Toggle

  • Proclaiming Themselves "The Good Guys," Believing Absolute Power Is a Necessary Evil
  • Military Contracts Spark Controversy, Hidden Safeguards Criticized
  • The Blind Spot of "Only I Can Solve It": The Danger of Lacking Checks and Balances

The development of artificial intelligence is an inevitable sprint, but on this path into the unknown, who has the right to hold the steering wheel? WIRED's latest in-depth feature sharply analyzes the "messiah complex" and corporate contradictions behind the renowned AI startup Anthropic.

Proclaiming Themselves "The Good Guys," Believing Absolute Power Is a Necessary Evil

Founded in 2021 by former employees who lost confidence in OpenAI CEO Sam Altman, Anthropic positions itself as a "Public Benefit Corporation." However, this company, which constantly warns that AI could trigger social unrest or even destruction, is now aggressively pursuing a valuation of nearly $1 trillion and actively seeking political and military influence.

The report notes that Anthropic's internal operations are based on two extremely confident beliefs: first, AI is the most transformational technology in human history, and its arrival is inevitable; second, to make the world better, Anthropic must maintain an absolute lead in the AI race.

Former employees reveal that the company often refers to itself internally as "The good guys," viewing OpenAI, Meta, and xAI as negative examples lacking responsibility. In internal talks, CEO Dario Amodei repeatedly emphasizes that the company must accumulate massive capital, computing power, and talent to generate enough "gravitational pull" to guide the entire industry toward safety. For them, the concentration of power is a "necessary cost" of fulfilling their salvific mission.

Military Contracts Spark Controversy, Hidden Safeguards Criticized

However, this strategy of "expanding by any means to achieve safety goals" has repeatedly triggered ethical red lines in practice. In 2024, Anthropic partnered with big data giant Palantir to provide AI services to U.S. intelligence and defense agencies. More recently, it was reported that the Pentagon used the Claude model to assist in identifying strike targets during the conflict between Israel and Iran.

When questioned about the potential for civilian casualties (e.g., an attack on an Iranian elementary school resulting in 120 deaths), Amodei only calmly responded that he was "unaware" and emphasized that as long as "the final decision is made by a human," it aligns with the company's safety policy.

Additionally, the newly released Claude Fable 5 model earlier this month has sparked major controversy. The model contains hidden safeguards that, if it detects researchers violating terms of service by using it for cutting-edge AI development (especially aimed at foreign competitors), will "secretly sabotage" their work progress. This move has drawn strong industry protests over overreach, ultimately forcing Anthropic to withdraw the backdoor and replace it with an open safeguard mechanism.

The Blind Spot of "Only I Can Solve It": The Danger of Lacking Checks and Balances

Regarding Anthropic's strategy, Helen Toner, former OpenAI board member and current CEO of Georgetown University's CSET, offered a vivid analogy: AI is like a forest filled with treasures and monsters, and everyone wants to rush in. Anthropic's approach is to "go the deepest first," dominate the negotiating table, and then try to tame the monsters.

But critics warn that this mindset of "only safe if we control it" has serious structural flaws. Shazeda Ahmed, a scholar at the University of California, Los Angeles (UCLA), points out that such organizations easily fall into a loop of "self-reinforcing beliefs," completely overlooking the blind spots of their own concentration of power.

Even though CEO Amodei recently acknowledged in an article the risk of "a small number of labs monopolizing AI power," his proposed solutions were "careful oversight" and "public commitments," rather than genuinely decentralizing power. WIRED concludes sharply at the end: No one truly knows how AI will change the world, but beneath Anthropic's idealistic veneer, we see only a tech behemoth that deeply believes it knows best and seeks to control all discourse.

View Original
This page may contain third-party content, which is provided for information purposes only (not representations/warranties) and should not be considered as an endorsement of its views by Gate, nor as financial or professional advice. See Disclaimer for details.
  • Reward
  • Comment
  • Repost
  • Share
Comment
Add a comment
Add a comment
No comments
  • Pinned