SE Gyges argues that AI safety’s public authority rests partly on a Berkeley social network whose funding, professional opportunities and intimate relationships overlap. The essay’s deliberately inflammatory title is “AI Safety Is Mostly A Sex Cult In Berkeley, California.” Its policy question is who gets presented as an independent expert, and what relationships that presentation leaves out.

Gyges follows connections among Eliezer Yudkowsky’s Machine Intelligence Research Institute (MIRI), the LessWrong community, effective-altruist philanthropy and frontier AI companies. Lighthaven, a Berkeley conference venue, becomes the essay’s physical center: a place where advocacy, recruiting and social life meet. Gyges argues that existential-risk beliefs can give insiders extraordinary moral authority while making dissent feel like betrayal of humanity.

The strongest evidence concerns specific institutional connections and failures:

  • MIRI’s own account of a 2013 strategy meeting identifies Dario Amodei and Holden Karnofsky as members of the effective-altruism community. It also records substantive disagreements with Yudkowsky. The connection is documented; uniform beliefs are a different claim.
  • The Center for Applied Rationality’s account of its handling of Brent Dill acknowledges serious vetting and safeguarding failures. It describes how an inadequate internal review could have given allegations the appearance of an impartial clearance.
  • Charlotte Alter’s reporting for TIME describes allegations of harassment, coercion and assault, alongside fears of professional retaliation. TIME says it corroborated accounts through other witnesses, contemporary records and screenshots, and includes the community organization’s response.

These sources make accountability a substantive issue. When jobs, housing, grants and relationships share a small network, reporting mistreatment can threaten several parts of someone’s life at once. Consensual polyamory itself establishes neither coercion nor misconduct. The relevant questions concern power, consent, conflicts of interest and independent routes for complaints.

Gyges goes much further, treating the network as the controlling center of essentially all AI safety. The essay also speculates about deliberate security incidents and a concealed violent agenda. Those allegations remain the author’s inferences; the cited connections and institutional admissions do not establish them.

There is an important qualification inside the essay itself. Gyges reproduces Rolling Stone’s account that an earlier allegation involving Yudkowsky was investigated and deemed baseless, and that the alleged victim rejected it. Gyges also withdraws an earlier version of this essay in its entirety, acknowledging that parts of its analysis exceeded what could be proved.

Read this as a polemic about concentrated influence, with documentary leads worth following individually. Its case for scrutinizing governance survives without accepting its claim that one social circle explains an entire research field.