Inside the lab, capability leadership and catastrophic risk warnings are the same strategy. Helen Toner's forest analogy names the logic critics now call stewardship through frontier position.
The analogy comes from Helen Toner, the Georgetown AI-governance researcher, in her conversation with WIRED. Picture a village on the edge of a vast forest, she tells the magazine. Inside the trees are magical treasures and dangerous monsters. Most villagers want to stay near the tree line, where the risks feel manageable. Anthropic, in Toner's telling, is the group that insists on going deepest, because only the explorer who reaches the center of the forest has any chance of taming what lives there.
It is a useful analogy, and it is exactly how Anthropic would describe itself. The lab has spent the last five years publicly warning that advanced AI could cause mass destruction, destabilize societies, or bring other grave harms. At the same time, according to WIRED, it is one of the most powerful forces pushing AI capabilities forward: a top developer and distributor of cutting-edge models, a courtship partner of the US military as a customer, and a company recently valued at almost $1 trillion. The risks Anthropic warns about are the very risks the company is racing to build into existence.
That is not hypocrisy, exactly. It is doctrine. Anthropic's two stated internal core beliefs, as reported by WIRED, run like this. First, AI is the most transformative technology in human history and its arrival is inevitable. Only the catastrophe-versus-prosperity branch is in question. Second, the world is better off if Anthropic remains at the frontier of the race. Several former employees, speaking on condition of anonymity, described that second belief to the magazine, and it underwrites almost everything the company does.
The structure is internally coherent. If transformative AI is coming regardless and the only question is who builds it, then capability leadership stops looking like recklessness and starts looking like responsibility. Compute investments, government contracts, and a near-trillion-dollar valuation are not departures from safety. They are the safety strategy. Inside Anthropic, leaders and employees have an internal shorthand for this stance, WIRED reports. Outside the company, observers describe the same logic as stewardship through frontier position. The further Anthropic pushes the capability frontier, the more resources, talent, and political leverage it accumulates, and the more credible its claim to be the steward of what comes next.
Critics say this is precisely where the doctrine breaks. Toner's forest analogy does the work for both sides. If the monsters can only be tamed at the center of the forest, then the explorer who reaches the center has the strongest claim to govern what is found there. But it is also unfalsifiable in real time. No one can run the counterfactual in which a different lab reaches the frontier first and handles the monsters better or worse. Anthropic's leadership is simultaneously the safety argument and the proof of it. Former employees told WIRED the posture struck them as self-serving: a convenient way to accumulate capital, compute, research talent, and political influence under a moral banner the company itself writes.
That critique is structural, not personal. It does not accuse Anthropic's leaders of bad faith, and the WIRED piece does not either. It points out that stewardship through frontier position is a self-reinforcing loop. Commercial success funds the frontier push, the frontier push produces the most capable models, and the most capable models justify the next round of commercial success. The doctrine cannot be tested against any external benchmark, because by construction the only company positioned to evaluate the doctrine is the company running it. Toner, in the same WIRED feature, is careful not to resolve the question for her readers. She names the forest and lets them decide how deep to walk.
Two things are worth watching. The first is whether Anthropic's government partnerships, including its courtship of the US military, stay inside the doctrine's own guardrails, or whether the company's working definition of "safe deployment" widens as the customer list grows. The second is whether the near-trillion-dollar valuation, as reported by WIRED, becomes a precedent that other labs feel obliged to match, pulling the entire frontier faster than any single safety argument can slow it.
A note on what this story is not. Anthropic has separately alleged that Alibaba used tens of thousands of fake accounts to copy Claude tens of millions of times. That is an intellectual-property fight, not a safety-doctrine fight, and the two share only a backdrop. Readers following both threads should keep them in separate folders.