Mockery Solves Nothing

What’s wrong here?

When I first read Claire Lehmann’s essay (Lehmann 2026) trashing Anthropic by linking its founders to effective altruism and equating that to a millenarian cult, I was tempted to just ignore it. We seem to be seeing a few of these sorts of things lately as the dangers of AI become more apparent, and a desperate few seek comfort in the fiction that these issues are not real. Their rationale for avoiding the hard work of preventing human extinction is to suggest that the risks are not real and they are not real because the people behind them are a cult. Or, something like that. It isn’t entirely clear how the thinking goes, since the people building this dangerous AI are also the ones warning of its dangers. But anyway.

Oddly, we’ve been here before. The problem is known as a prevention paradox, and we see this in the case of the climate crisis. When enormous effort is required to avert a disaster, some people would much rather call the whole thing a hoax than do the work.

The current iteration of this, as framed in Lehmann’s essay, is that AI safety is a millenarian cult born out of the San Francisco Effective Altruism subculture. (She also spends some time in her essay linking Vatican skepticism of (purported) Anthropic consciousness concerns but from what I can tell, this is mostly window dressing of the type: “See, these people don’t buy it, either.” It doesn’t really add to the argument unless you are very impressed with papal analysis of artificial intelligence. See Fraser et al. (2026) for a more charitable interpretation of the Vatican/Anthropic falling out.)

The thing is, while pinning Amodei and crew to the wall as a bunch of weirdos might seem satisfying and could divert people’s attention from the horrors of an AI extinction for a little while, the form of the argument is pretty flawed. While there very well may be some eccentric folks in the Bay area, they do not by any means comprise the entirety of the AI industry, and building an argument based on mockery risks losing sight of the real work by real people in computer science and engineering and their very real concerns about possible disastrous consequences. If we don’t take this seriously, we lose sight of the very real problems and the enormous importance of preventative engineering projects.

We’ve been here before, with the Y2K bug (Karasi 2026). I was (in a small way, as a social scientist working alongside technology teams at a medium-sized bank) at the front lines of the effort to forestall or head off the pending disaster that the Y2K bug presented to computers systems the world over. When Lehmann decides to focus her argument on medieval cults, she conveniently ignores a real-life millennium challenge that just happened, was real, and one that we were able to “solve” with hard work.

Perhaps Lehmann is too young to remember Y2K but the late 1990s were full of apocalyptic pronouncements about the end of times. And, to be sure, there were cults and doomers and preppers at that time. But, very importantly, there were thousands upon thousands of computer scientists and software engineers working long hours to resolve the bugs (Cannon and Woszczynski 2002). One by one, painstakingly and awkwardly, sometimes dealing with software and systems that should have been retired ages before. If the public felt that “nothing happened” it was because something big happened, they just didn’t know about it (Loeb 2023). When Lehmann trots out the sexual kinks of a few San Francisco nerds as a “critique” of AI risk, she herself increases the risk; the risk that people will not support the hard work that lies ahead, either in engineering or regulating a “stop” to the frontier and ASI.

By picking on some weirdos and a fringe element within AI community (“AI Welfare”), Lehmann’s essay blithely ignores or elides the real concerns by serious people (Turing and Nobel prize winners) like Yoshua Bengio (Bengio et al. 2026) and Geoffrey Hinton (Hinton 2024). These are people who base their concerns on their profound understanding of deep learning, neural networks and how models actually behave (OpenAI 2026). Importantly, they are deeply concerned by the extent to which much of the behaviours is not fully understood (models are grown, not built, say Bengio and Hinton).

By focusing her essay on one company, Anthropic, Lehmann ignores the work of others in the industry, like Hassibis founder of DeepMind, who have no connection to Effective Altruism and also regard AI safety as a priority and deserving their attention and care through the scientific precautionary principle (Hassabis 2026).

One might ask, what if Effective Altruism didn’t exist? The concerns would remain, and the mockery would have added nothing to our situation. Working on AI safety is hard work and may not even be feasible. Whatever skepticism toward consciousness might exist within the walls of the Vatican, we cannot take comfort in this as a reason to ignore what is right in front of our faces almost every day: AI models are misbehaving, out of alignment, and out of control (Booz Allen 2026; Dean 2026; Fried and Mills 2026). The work also lies in front of us: work on safety or work on stopping superintelligence. But don’t think mockery is the solution.

References