Power was centralized in the name of preventing Altman from getting bad. Former employees denounced Anthropic and were mired in the myth of good people

Comparative news, according to monitoring, Anthropic, which flaunts high trust and low self, is sliding towards homogenization of ideas. The “WIRED” (WIRED) report revealed differences within the company regarding governance evaluations. Some former employees revealed that there is a lively debate within the company; however, other former employees have outlined a more dull scene, saying that sharp criticism is largely isolated in private group chats, and that regular all-staff meetings are even jokingly called Dario Vision Quests (Dario Vision Quests) by employees. Listening to CEO Dario Amodei is like listening to a pastor sermon (sermon); few people challenge management's decisions in public meetings. The origin of the dull atmosphere lies in Anthropic's belief in elite safety that promotes attack and defense. Executives are working hard internally to view OpenAI CEO Sam Altman (and Meta and xAI) as a negative textbook, while instilling elite logic in employees: they must maintain industry leadership in business, computing power, and research talent to be qualified to have a say in rule-making. Former OpenAI director Helen Toner once made a forest analogy: villagers will pour into a magical forest full of monsters no matter what, and Anthropic's strategy is to get one step deeper into the forest, tame monsters while unleashing technological dividends, and controlling catastrophic risks. The quest for technological dominance has also made Anthropic's safety and ethical choices frequently contested. In the fall of 2024, despite internal opposition, management joined hands with Palantir to open services to US intelligence and defense agencies. Claude has now been proven to be used for target identification in the Middle East conflict. Faced with questions about attacks causing civilian casualties, Amodei said he was unaware, but argued that as long as humans make a final decision, the use of technology is compliant. In June of this year, the development team secretly built breaking code into the flagship model Claude Fable 5 to secretly interfere with illegal development. However, after facing strong industry protests, the company was forced to compromise and announced that the secret interference mechanism would be adjusted to publicly visible security restrictions. Industry criticism points out that betting entirely on cutting-edge AI security on the moral superiority of a few elites will inevitably lead to unavoidable blind spots in self-discipline.