Additional Coverage:
- The Anthropic–OpenAI feud and their Pentagon dispute expose a deeper problem with AI safety (fortune.com)
AI’s Power Struggle: A Tech Titan Tussle Raises Red Flags for Safety
Local Angle: While the national spotlight often shines on the glitzy advancements of artificial intelligence, a recent feud between two tech giants, Anthropic and OpenAI, is casting a shadow on the very foundation of AI safety, raising uncomfortable questions about who truly controls this powerful technology.
The ongoing drama, highlighted in a leaked memo from Anthropic CEO Dario Amodei, reveals a bitter rivalry that goes beyond mere corporate competition. It underscores a deeper concern: the concentrated power of a few corporate leaders and government officials in shaping the future of AI.
For years, critics have warned about “industrial capture,” where a handful of companies, working hand-in-hand with governments, dictate AI development. This week’s events seem to be a stark realization of those fears.
The recent flare-up between Amodei and OpenAI CEO Sam Altman ignited after OpenAI secured a deal to provide AI to the Pentagon, leading Secretary of War Pete Hegseth to declare Anthropic a “supply chain risk” for failing to secure a similar agreement. Amodei’s internal memo didn’t pull any punches, lambasting OpenAI’s messaging as “mendacious” and “safety theater,” and accusing Altman of “straight up lies” and “gaslighting.”
Altman, for his part, has publicly returned fire, calling Anthropic’s Super Bowl campaigns “clearly dishonest” and accusing them of “doublespeak.” The animosity even extended to a recent summit, where the two CEOs famously refused to hold hands for a group photo with Prime Minister Narendra Modi – a small but symbolic gesture of their deep-seated rivalry.
With minimal government regulation and stalled international efforts on AI safety, the world has largely relied on the industry to self-regulate. Both OpenAI and Anthropic have publicly supported this approach and signed voluntary safety commitments, even collaborating on independent safety evaluations. However, the escalating animosity between their leaders raises a critical question: how much genuine cooperation on safety can we realistically expect when competition is this fierce?
The pressure to compete has already impacted both companies’ commitment to safety. Anthropic recently revised its Responsible Scaling Policy, indicating it would no longer unilaterally delay developing a new model simply because it hadn’t yet figured out how to make it safe.
OpenAI has also made shifts, removing explicit bans on military and warfare uses from its policies and re-prioritizing product development over safety research. This shift prompted former OpenAI superalignment lead Jan Leike (who has since moved to Anthropic) to publicly state that at OpenAI, “safety culture and processes have taken a backseat to shiny products.”
The current approach to AI safety operates on the assumption that companies and governments will act with restraint. However, the future of AI safety may ultimately hinge on how a small group of powerful players navigate the complex pressures of competition, geopolitics, and, it seems, the occasional Silicon Valley soap opera.