OPENAI VS ANTHROPIC: THE AI SAFETY WAR
Dario Amodei was OpenAI’s VP of Research.
He led the safety team.
In 2021, he quit.
Reason: Sam Altman was prioritizing deployment over safety research.
Dario took a dozen top OpenAI researchers with him and started Anthropic.
The pitch to investors was simple: “We’re the responsible AI company.”
Constitutional AI. Harmlessness training. Ethical alignment.
They raised $7 billion in 18 months. Google invested. Amazon invested. Governments took meetings.
OpenAI kept releasing faster models.
Anthropic promised to move carefully.
Then both companies took Pentagon contracts.
OpenAI signed deals for military decision-support systems.
Anthropic signed deals for intelligence analysis and data processing.
The “safe” AI company and the “fast” AI company both decided military funding was too valuable to refuse.
Congress called hearings on AI safety.
Sam Altman testified. Promised OpenAI would prioritize safety.
Dario Amodei testified. Said Anthropic was already doing that.
Then OpenAI released GPT-4o.
Anthropic released Claude 3 Opus.
Both more powerful. Both faster. Neither slowed down.
Here’s what nobody says:
Dario didn’t leave because he thought AI should be slower.
He left because he wanted to be the one controlling how fast it goes.
Anthropic’s “Constitutional AI” isn’t safer. It’s just different constraints.
OpenAI’s “move fast” isn’t reckless. It’s just transparent about priorities.
Both companies are in the same race.
Both took military money.
Both ignored their own warnings.
The only difference? One has better PR about safety.
Wow, reading this really put things into perspective about the so-called 'AI safety war' between OpenAI and Anthropic! I always saw Dario Amodei's departure from OpenAI as this clear-cut stand for safety, thinking Anthropic was truly going to be the slower, more cautious player. But the revelation about both companies, even Anthropic, taking Pentagon contracts? That really hit home. It makes you wonder if 'AI safety' is sometimes more about clever branding and PR than actual practice, as the article suggests. I've been following the concept of 'Constitutional AI' that Anthropic champions, aiming for ethical alignment and harmlessness training. On the surface, it sounds incredibly promising – a way to embed values directly into the AI's core. It's a stark contrast to OpenAI's 'move fast and break things' ethos, or at least how it's often perceived. Yet, after seeing that both ended up in the same race, rapidly deploying powerful models like GPT-4o and Claude 3 Opus, it leaves me questioning how different their fundamental approaches truly are when faced with market and competitive pressures. Is 'Constitutional AI' just a different set of constraints, as the article puts it, ultimately leading to similar outcomes? The military funding aspect is particularly unsettling for me. The idea of advanced AI from either the 'safe' or 'fast' camp being used for military decision-support systems or intelligence analysis raises so many ethical questions. It highlights the immense power and dual-use nature of this technology. It feels like a moment where the initial promises of responsible AI development get tested by real-world incentives. When billions of dollars are on the table, does the commitment to 'harmlessness' or prioritizing safety over deployment genuinely hold up? It seems like both Sam Altman's vision and Dario Amodei's original intent might have been swayed by the sheer scale of the AI arms race. This whole situation really underscores how complex AI development is. It’s not just about technical breakthroughs but also about navigating intense competition, investor demands, and geopolitical pressures. It makes me think that simply having a 'safety team' or a specific philosophical approach might not be enough. What we really need is transparency and constant scrutiny of how these powerful tools are being built and deployed. It's a reminder that we, as users and observers, need to look beyond the headlines and PR spin to understand the true dynamics at play in this critical field. It's a race, for sure, but the finish line might look very different from what we initially imagined.