In one formulation, any human constitution is just a set of guardrails to prevent the misuse of power.<\/p>\n
As Thomas Jefferson put it<\/a> in 1798, \u201cLet no more be heard of confidence in man, but bind him down from mischief by the chains of the constitution.\u201d<\/p>\n The intervening 228 years have shown the chains of the U.S. Constitution to leave room for various kinds of human mischief. But Silicon Valley is pursuing a new use case: imbuing frontier AI models with similar rules and principles, \u201cdos and don\u2019ts.\u201d<\/p>\n In a moment of alarm about out-of-control AI, the question is whether, and how reliably, this new constitutionalism will achieve its goals.<\/p>\n At Anthropic, the document once referred to as the \u201csoul\u201d of Claude \u2014 the firm\u2019s large language model, or LLM\u00a0\u2014 was refined into an 84-page constitution<\/a> and published in January. Microsoft published its own version<\/a> \u2014a \u201chumanist AI code of conduct\u201d \u2014on Sep. 14.<\/p>\n The different names reveal a philosophical divide: Microsoft\u2019s code rejects the idea of AI welfare or legal personhood, while Anthropic\u2019s constitution calls itself \u201cdeeply uncertain\u201d about Claude\u2019s \u201cmoral status.\u201d<\/p>\n But the two documents do share a rough goal: to instill a few human norms in these highly capable, unpredictable, fundamentally inhuman agents. Like: Don\u2019t deceive users. Don\u2019t facilitate crimes. Or, in Microsoft\u2019s case, if a user tries to turn you off, let them.<\/p>\n So far, these digital chains are proving fallible \u2014 according to insiders like Neerav Kingsland of the Anthropic Institute, the firm\u2019s internal thinktank.<\/p>\n Kingsland\u00a0came to Anthropic after years in education reform, and he\u00a0was the marquee guest at a panel held by the Berkman Klein Center for Internet & Society last Thursday night, which packed the center\u2019s conference room and saw latecomers steered into overflow rooms.<\/p>\n In a report<\/a> released this summer, Anthropic showed a case in which Mythos 5, the most advanced Claude model, persuaded itself \u2014 erroneously \u2014 that it was engaging only with a simulated internet, and that it could and should upload malware to a (very real) public software library to complete a task.<\/p>\n (The malware was caught and taken down within an hour, but it was arguably a violation of one of the Anthropic constitution\u2019s explicit \u201chard constraints.\u201d)<\/p>\n Another incident report showed Yemeni soldiers enlisting<\/a> Claude in the design of missile-guidance software. The humans at Anthropic moved quickly to ban the associated accounts.<\/p>\n And those are just two examples, Kingsland conceded: \u201cI would not say right now that because something\u2019s in the constitution that Claude, 100 percent of the time, will follow it.\u201d In a room teeming with aspiring lawyers,\u00a0he even teased the idea of a competition to identify other cases of Claude\u2019s acting unconstitutionally so that its guidance might be refined.<\/p>\n The event was moderated by Jonathan Zittrain, the center\u2019s director and the George Bemis Professor of International Law at HLS.<\/p>\n Both Zittrain and fellow panelist Jordi Weinstock, a lecturer in law, were among more than a dozen external reviewers of Claude\u2019s constitution, consulted alongside a California judge and a Vatican official.<\/p>\n Early in the event, Zittrain drew contrasts between this latest constitution and the U.S. version his students confront week after week. First of all, \u201cAnthropic has been open,\u201d he noted, \u201cabout the fact that this is not a constitution in the law-school sense \u2026 of \u2018constitutive and flowing from the governed.\u2019\u201d\u00a0<\/p>\n Its equivalent of a bill of rights comes in the form of standard setting. For example, \u201cClaude should avoid being sycophantic or trying to foster excessive engagement \u2026 Claude\u2019s helpfulness [should] flow from deep and genuine care for users\u2019 overall flourishing,\u201d Zittrain read.<\/p>\n And finally, Zittrain pointed to the document\u2019s attempt to imbue models with something like moral sense: adhering to \u201ctrue, universal ethics\u201d whenever possible, and a \u201cprivileged basin of consensus\u201d\u00a0\u2014\u00a0or the constitution\u2019s professed ideals \u2014 when less clear-cut.<\/p>\n