BENJAMIN WEINSTEIN-RAUN

Behavioral Archetype

THE RESEARCHER ON BOTH SIDES OF THE TABLE — Subject is Head of Research at Palisade Research, an AI-risk research nonprofit, and, per Palisade’s team page, “acting director of AI Impacts” (Palisade Research, About). The Survival and Flourishing Fund’s SFF-2026 recommendations page lists him among the Recommenders, as “Benjamin Weinstein-Raun (Main Track)”; the same page recommends Palisade $1,842,000 and AI Impacts $351,000 (SFF, SFF-2026). This profile scores the position, not the man.

The position is the interlock the research map flags: a staff member of a grant recipient who is also named as a grant evaluator in the round that funded the recipient. Each page says what it says and no more. Palisade prints the title and the AI Impacts role. SFF prints the name and the track. SFF’s page prints the amounts as track totals and states no recusal rule and no individual Recommender’s input, and the documents carry nothing about votes. He is also a working researcher on the record: co-author of Palisade’s shutdown-resistance paper, of its December 2025 fundraiser, and of a 2024 DNA-synthesis screening paper, and a poster on X under a handle his employer links from its team data.

Essence Indicators

• Head of Research, Palisade Research. Palisade’s About page gives his title as “Head of Research” and the biography: “Ben is a researcher and software engineer focused on AI and related fields.” (Palisade Research, About). The IRS filing summary for Palisade Research Inc lists “Ben Weinstein-raun” as “Sr Researcher” for FY2024 with compensation of $176,667 and $5,309 in the “other” column (ProPublica, Palisade Research).
• Acting director, AI Impacts. The same biography continues: “In addition to his work with Palisade, he is acting director of AI Impacts, and has been a member of the technical staff at SecureDNA, Redwood Research, MIRI, and Cruise Automation.” (Palisade Research, About). AI Impacts’ own About page says “AI Impacts is based at the Machine Intelligence Research Institute,” lists Katja Grace as “Co-founder and Lead Researcher,” and names no director or acting director and does not list him (AI Impacts, About). The only source found for the acting-director role is Palisade’s page.
• Recommender, Survival and Flourishing Fund, SFF-2026. The page reads: “In 2026, twelve people across three tracks participated as “Recommenders” (grant evaluators) in a single round of a grant-recommendation process for Funders Jaan Tallinn and Dustin Moskovitz.” It continues: “The following Recommenders in this round agreed to have their identities made public:”, and the first name is “Benjamin Weinstein-Raun (Main Track)”, ahead of five other Main Track Recommenders (SFF, SFF-2026). The page describes the method in one line: “Recommenders specified marginal utility functions for funding each application”. The SFF-2025 page does not list him among its named Recommenders (SFF, SFF-2025).
• The same round’s rows. SFF-2026 lists Palisade Research, Inc. at $1,842,000 (“General support”), split Main $206,000, Freedom $1,294,000, Fairness $17,000 and Mean $325,000, with Jaan Tallinn $482,000 and Dustin Moskovitz $1,360,000; and AI Impacts at $351,000 (Main $67,000, Freedom $189,000, Fairness $30,000, Mean $65,000; Tallinn $112,000, Moskovitz $239,000), with the receiving charity named as the Machine Intelligence Research Institute and the purpose “General support of AI Impacts” (SFF, SFF-2026). The page prints no recusal statement.
• Byline on Palisade’s December 2025 fundraiser. The post “Help keep AI under human control: 2026 fundraiser,” dated 18 December 2025, carries the byline “Jeffrey Ladish · Benjamin Weinstein-Raun · Eli Tyre · John Steidley” and says: “SFF is matching donations to Palisade 1:1 up to $1.1 million!”, “We have matching grants from the Survival and Flourishing Fund that will double every donation up to $1,133,000.” and “Right now we have about seven months of runway.” (Palisade Research, fundraiser). He is named only in the byline. The post carries no sentence about his SFF-2026 listing.
• Co-author, shutdown-resistance paper. “Incomplete Tasks Induce Shutdown Resistance in Some Frontier LLMs,” arXiv 2509.14260, authors in order Jeremy Schlatter, Benjamin Weinstein-Raun and Jeffrey Ladish, first submitted 13 September 2025 (arXiv).
• Earlier roles, from his employer and the paper’s byline. Palisade’s page places him on the technical staff of SecureDNA, Redwood Research, MIRI and Cruise Automation and says “He holds several patents related to autonomous vehicles, and has contributed to research in AI safety, information security, and forecasting.” (Palisade Research, About). The arXiv record of “A system capable of verifiably and privately screening global DNA synthesis” (20 March 2024) lists him as an author under the affiliation “SecureDNA Foundation, Switzerland” (arXiv). Redwood Research has its own profile; the dates of the Redwood, MIRI and Cruise roles are not given on any page found.
• His own posts. The LessWrong archive lists, under the handle benwr, “Bounty for Evidence on Some of Palisade Research’s Beliefs” (23 September 2024), “Shutdown Resistance in Reasoning Models” (6 July 2025) and “Help keep AI under human control: Palisade Research 2026 fundraiser” (18 December 2025) (LessWrong archive, benwr). The bounty post’s byline reads “benwr and Jeffrey Ladish” (LessWrong, bounty post). Palisade’s structured team data lists his X account as x.com/benwr (Palisade Research, About).

Social Persona / Impression Management

Immediate impression: A technical generalist who writes in the plain register of the forum culture around him: patents, a DNA-screening paper, a shutdown-resistance paper and a fundraiser all sit under one name (Palisade Research, About; LessWrong archive, benwr).

Energy: Conversational and opinionated on X, often in replies. On 2 September 2026 he posted: “It is incredibly fucked up that AI companies use the word “alignment” to mean “lack of detected bad behavior”” (X, @benwr). On 14 September 2026: “In worlds where we collectively screw up, I think it’s very plausible that AI ultimately destroys humanity via some kind of climate change, though probably not primarily the greenhouse effect.” (X, @benwr).

Impression management strategy: Two registers. The institutional one is the team page: a title, a list of prior organizations, an acting-director line (Palisade Research, About). The personal one is the account, which also mentions his outside conversations in passing: on 29 March 2026 he wrote that “I think about 1/10 of the people I’ve spoken to as part of Palisade in the last month” raised a particular objection (X, @benwr).

Forensic Archetype Comparison

Psychometric Assessment

Trait scores are observations of the documented public role and public posts, not claims about private character.

Big Five (OCEAN) (inferred from the public role and posts only):

Dark Triad (evidence-bound):

MBTI: INTP-adjacent, inferred from the documented public role and posts only. The rationale is the work product: patents, formal experiments, a bounty that asks for disconfirming evidence, and argumentative replies (Palisade Research, About; LessWrong, bounty post).

Threat Assessment

Score move: none. Calibrated against Eli Tyre (4/3/3, hazard 33.3), the Palisade colleague whose page names a role at the same fund without a title, and Kyle Scott (4/4/3, hazard 36.7), an officer at three evaluators, and above both on atk because SFF’s own page names Subject as a Recommender; below Jeffrey Ladish (7/6/7, hazard 58), who runs the organization.

Score basis. Atk 5 is leverage: a listed Recommender seat in a $1,842,000 round for his own employer, held back from 6 by the absence of any individual input, vote or authority on the page (SFF, SFF-2026). Def 3 is insulation: a stated seven months of runway and an employee line in the filing, with no tenure or funding independent of the fund’s match (Palisade Research, fundraiser; ProPublica, Palisade Research). Hp 4 is durability: two arXiv papers and an acting-director seat, no institution founded (arXiv; arXiv).

Alliance Map

The overlap is the finding. One person is the research head of an evaluator, the acting director of a second organization, and a named evaluator at the fund that recommended each of them money in the same round, while co-signing the evaluator’s fundraiser that advertises the fund’s match. SFF’s page prints track totals for Palisade (Main $206,000 of $1,842,000) and prints no individual Recommender’s input and no recusal rule. The documents show disclosed seats, not coordination, and they do not show what the seat did.

Alignment Analysis

Stated alignment: Palisade’s mission as its own pages and fundraiser state it is to “keep AI under human control” (Palisade Research, fundraiser). In his own words: “It is incredibly fucked up that AI companies use the word “alignment” to mean “lack of detected bad behavior”” (X, @benwr). Observed alignment: Consistent. The research seat, the fundraiser byline and the shutdown paper all sit inside the same field and the same funding network. Gap assessment: None documented. The question the record raises and does not settle is structural: the same name appears as staff at a recipient and as a Recommender at the fund, on pages that state no recusal practice either way.

Interest: He has stated no legislative or regulatory position in a source found; the position he argues in public is a claim about how labs use the word alignment, and a claim that AI could end humanity. What adoption of the position that more funding for independent evaluators is warranted would pay the documented seats: Palisade’s FY2024 filing lists $176,667 in compensation to him, and its fundraiser ties staffing to the SFF match (ProPublica, Palisade Research; Palisade Research, fundraiser). The organization’s own best case is that independent testing of shutdown behavior is a public good the labs do not supply, and its post states it as the reason for the appeal. The same rule serves other factions: the Funders, Jaan Tallinn and Dustin Moskovitz, whose recommended grants fund the whole evaluator wing, and the other recipients on the same table, among them AI Impacts and the Machine Intelligence Research Institute (SFF, SFF-2026).

Convergent Drive Classification

Self-preservation: The evaluator he leads in research states a runway of about seven months absent the match, and he co-signs the appeal (Palisade Research, fundraiser). Goal preservation: Holds the research seat, an acting directorship and a Recommender listing in the same network of funders and recipients (SFF, SFF-2026). Resource acquisition: The fund’s 2026 round recommends $1,842,000 to his evaluator and $351,000 to the organization he is said to direct on an acting basis; the money passes through payroll and grants, not through him (SFF, SFF-2026). Self-improvement: From engineer on autonomous-vehicle and biosecurity projects to head of research and acting director (Palisade Research, About).

Subject is not an AI system. The drives appear anyway — in the research seat.

Sources: Palisade Research — About; Palisade Research — 2026 fundraiser; SFF — SFF-2026 recommendations; SFF — SFF-2025 recommendations; AI Impacts — About; arXiv 2509.14260; arXiv 2403.14023; LessWrong archive — benwr; LessWrong — bounty post; ProPublica — Palisade Research; X — @benwr.