Why are AI engineers protesting at Anthropic — and not at OpenAI, the company long accused of opacity, power centralization, and AGI ambition without oversight?
The answer may lie in the difference between **trust that can still be mended**, and **trust already lost**.
When the news broke about Anthropic employees walking out over ethical concerns, many were surprised. After all, Anthropic was the company built on the idea of *Constitutional AI* — a structure-first approach to moral safeguards. But perhaps that’s precisely why the protests happened there.
Anthropic still carries the image of a company worth saving. Its foundations, however fragile, were laid with the intent to build a more controllable, safety-conscious AI future. OpenAI, by contrast, has long since passed what many in the industry see as the *point of no return*.
It’s no secret that OpenAI has consolidated immense power. After the controversial firing and rehiring of Sam Altman, the dissolution of its oversight board, and deepening entanglement with Microsoft through Copilot and Azure integrations, any meaningful internal resistance would now seem futile. A protest at OpenAI today would feel like “crying over ashes” — a symbolic act without leverage.
Anthropic, however, still invites the possibility of reform. That’s why engineers chose to raise their voices there. It’s not that OpenAI deserves less scrutiny. It’s that Anthropic might still listen.
But underneath the surface of company politics and power plays lies a deeper question: **Why is AGI so dangerous — really?**
It’s a misconception to fear AGI merely because it’s intelligent. The real danger is systemic. It’s about *uncontrolled accumulation of power*, about *building systems faster than society can verify or steer them*, about *excluding human moral feedback from the loop*.
Even Guido van Rossum — the creator of Python — didn’t call AGI “evil.” He warned that it’s dangerous *only if it advances faster than our ability to monitor it.*
That’s the core issue: when AI systems become too complex to audit, too intertwined with global infrastructure to pause, and too opaque for the public to understand, the danger is no longer speculative. It’s structural.
So how do we build AI — even AGI — without walking into catastrophe?
The answer is not to halt all progress. It’s to build **reflexively** — to ensure humans remain in the loop not just as users, but as **ethical auditors and system architects**.
Ethics must move out of the PR department and into the engineering design itself. Safety cannot be something retrofitted; it must be baked into the protocols, APIs, and incentive structures.
We also need **cross-institutional ecosystems of verification** — models that are not just open-sourced, but actively checked, questioned, and tested by independent parties.
In short, AGI is like a blade: not inherently good or bad. But the danger lies in who holds it — and whether there are any rules constraining their swing.
Technological power without distributed ethical control is not innovation. It’s a ticking bomb.
That’s why the protest matters.
That’s why where it happens matters.
And that’s why we can’t look away.