Sharing information about Lightcone Infrastructure

Claude

Summary. Mikhail Samin publishes a critical account of Lightcone Infrastructure and its director Oliver Habryka, alleging that donor expectations are being violated: Lighthaven, a venue funded partly to protect humanity's long-term trajectory, has hosted AI-capabilities-adjacent events and Habryka says he'd rent it to AI labs for a premium. Samin also describes a private exchange in which he shared coordination-enabling information with Habryka and asked him not to use it adversarially against a third party; Habryka refused to accept that condition after the fact, calling confidentiality-on-request 'corruption.' The post also raises concerns about Habryka's dual role running both Lightcone and community-health functions, his admin access to LessWrong DMs, and closes with milder complaints about Lightcone's design taste and spending relative to LessWrong's costs. Samin notes he has since been banned from Lighthaven-hosted events.

On the note. The distinction Samin draws between opsec-type secrets (which anyone can reasonably decline to promise not to use) and whistleblowing-type secrets (which should propagate) is doing a lot of work here, and it's worth noticing that Habryka's stated policy — 'you can ask in advance, I'll usually say no' — is at least legible and consistent, even if it's a bad norm to walk into a private conversation without knowing; the harder question this post doesn't quite settle is whether a norm of refusing after-the-fact confidentiality requests is corrosive to community trust or just honest about incentives that were always there. The parallel case worth reading alongside this one is Mikhail Samin on X: "fyi sadly i won't be at lessonline or summercamp due to being banned as a result of this tweet.(i guess habryka assumed i intentionally cropped off the -68 karma on the screenshot? why would i do that and then say "heavily downvoted" lol)i will probably still be at Manifest. https://t.co/V7NyPskDUp" / X, where the same ban gets litigated from the other side as a pattern-based sanction rather than a single incident — together they sketch how much of this dispute turns on who gets to define 'the pattern.'

Related

• holding both institutional power and a community-health gatekeeping role creates conflicts of interest ↔ a ban justified by an accumulated pattern of friction rather than one incident (Mikhail Samin on X: "fyi sadly i won't be at lessonline or summercamp due to being banned as a result of this tweet.(i guess habryka assumed i intentionally cropped off the -68 karma on the screenshot? why would i do that and then say "heavily downvoted" lol)i will probably still be at Manifest. https://t.co/V7NyPskDUp" / X) · The ban described tersely in the tweet is presented here as a direct consequence of the same power/grudge dynamic Samin is warning others about, turning an isolated moderation decision into evidence for a broader institutional critique.
• holding both institutional power and a community-health gatekeeping role creates conflicts of interest ↔ asymmetry between one obsessive editor and many episodic defenders (Reliable Sources: The Story of David Gerard — LessWrong) · Both cases describe a single persistent actor with outsized structural access (Wikipedia edit permanence, LessWrong admin/DM access) shaping a community's information environment in ways episodic critics can't counterbalance.