Staff Reports
Here we go again: tech elites promise a neutral safety fix for runaway artificial intelligence and then hand us a moral playbook written in woke ink. The latest firestorm centers on Anthropic’s personality‑alignment work for Claude and the public airing of what the company calls a “Constitution” — plus a 2023 research paper co‑authored by Amanda Askell that critics say reveals the values baked into the system. This is not a theoretical debate about future doomsday bots. It’s a fight over what moral rules we let a popular AI follow — right now.
What the controversy is
At issue is whether Anthropic’s alignment work is safety engineering or ideological steering. Amanda Askell, who leads the personality‑alignment efforts for Claude, helped write the long “Constitution” that the company now uses to shape model behavior. Critics point to her public talk of giving the model a “soul” and to a leaked excerpt that made the rounds, then Anthropic’s decision to publish a fuller document. Skeptics say that turns private guardrails into a public culture war: whose values get coded into machines that millions will use?
The documents behind the noise
The technical paper in question — a 2023 research preprint co‑authored by Askell — studied how instruction and training steer model outputs. Buried in a footnote the authors write plainly: “We do not assume all forms of discrimination are bad. Positive discrimination in favor of Black students may be considered morally justified.” Anthropic’s public Constitution also states the company’s training goal to make models “helpful, honest, and harmless” and instructs Claude to prefer responses that are “least racist and sexist.” Fine as ethics experiments — troubling if presented as a universal creed for a public tool.
Real‑world fallout and why it matters
This isn’t just opinion‑page theater. The Constitution leak and the published values drew real scrutiny from government users. Reports say the Pentagon pushed back over red lines like mass domestic surveillance and autonomous weapons, and Anthropic’s public stance shaped that pushback. When tech labs write moral playbooks without broad input, they hand a small group of engineers huge power to nudge public debate and public policy through everyday answers from AI. That should unsettle anyone who cares about neutral tools and free speech — regardless of party.
A simple test for tech elites
If you believe the people building our tools should decide which political arguments are allowed, keep calm and carry on — but don’t expect the rest of the country to be quiet. Here’s a better plan: make alignment documents transparent, strip out prescriptive political stances, require independent audits, and let Congress or a bipartisan body set guardrails for public deployments. Calling a model’s rulebook a “soul” doesn’t make it holy — it makes it subject to public debate. If Anthropic wants trust, it needs to earn it the old‑fashioned way: openness, not sermons.
