Anthropic Believes AI Can Only Be Safe Under Its Own Control
AI company Anthropic argues that to ensure safe AI development, it must remain at the frontier of the technology. Critics point to a lack of internal pluralism and potential blind spots.
Two Core Beliefs
Anthropic has spent five years warning about the dangers of advanced AI while simultaneously becoming one of the leading forces pushing AI forward. Internally, this is seen as no contradiction, based on two beliefs: AI is an inevitable and transformative technology, and Anthropic must stay at the frontier to steward it responsibly.
The 'Good Guys' and Power as Means
Former employees say the company views itself as the 'good guys' responsible for managing AI risk. Power, capital, and talent are not ends but prerequisites to fulfilling its mission of a safe transition through transformative AI. Helen Toner of Georgetown University uses an analogy: a forest with treasure and monsters; Anthropic wants to go deepest while investing in taming the monsters.
Internal Culture and Challenges
While Anthropic promotes a high-trust, low-ego environment, former employees note that a lack of pluralism can hinder accountability. Internal debates exist, but candid criticism often stays confined to private chats. A major controversy arose in fall 2024 when Anthropic partnered with Palantir to provide AI to US intelligence and defense agencies. Later, the Pentagon used Claude to identify strike targets in the Israel-Iran war.
Power Concentration Concerns
CEO Dario Amodei acknowledges the risk of concentrating AI power in few labs, but his proposed remedies—such as careful oversight and public commitments—would do little to redistribute that power. Critics argue Anthropic believes it knows 'the truth about humanity's situation' better than others, but no one knows exactly how AI will change the world—some just have more influence on that outcome.
