JARED KAPLAN

Behavioral Archetype

THE OFFICER OF THE POLICY HE HELPS REWRITE — Subject is the physicist who measured how model loss falls with scale, then co-founded the lab that sold the curve, then took the seat that signs off on releases under the lab’s own safety policy. The record has three layers. He is first author of “Scaling Laws for Neural Language Models” (arXiv:2001.08361). He is named Responsible Scaling Officer: “Co-Founder and Chief Science Officer Jared Kaplan will serve as Anthropic’s Responsible Scaling Officer, succeeding Co-Founder and Chief Technology Officer Sam McCandlish” (Anthropic, October 15, 2024). And the policy says “Changes to the RSP will be proposed by the CEO and RSO,” so the officer who applies the rule also proposes it (Anthropic, RSP v3.0; v3.4). In February 2026 he explained to TIME why the unilateral commitment was removed. This profile scores the seat, not the man.

Essence Indicators

• Co-founder and Chief Science Officer, Anthropic. The October 15, 2024 post gives both titles (Anthropic). The Guardian describes him as “a Stanford-and Harvard-educated professor of physics who researched at Johns Hopkins University and at Cern in Switzerland before joining OpenAI in 2019 and co-founding Anthropic in 2021” (The Guardian, December 2, 2025). The Johns Hopkins department page lists him as “Associate Professor” with “Research Interests: Quantum Gravity, Conformal Field Theory, Machine Learning,” and states no leave (JHU Physics & Astronomy). The Company page’s lists of directors (Dario Amodei, Daniela Amodei, Yasmin Razavi, Reed Hastings, Chris Liddell, Vas Narasimhan) and Long-Term Benefit Trust trustees (Ben Bernanke, Richard Fontaine, Neil Buddy Shah, Christopher Stubbs) do not include him (Anthropic, Company).

• Responsible Scaling Officer: the seat and its powers. The policy says: “We will maintain the position of RSO, a designated member of staff who is responsible for the implementation of this policy. The RSO’s duties will include (but are not limited to): (1) as needed, proposing updates to this policy; (2) approving relevant model development or deployment decisions based on our risk assessments; (3) reviewing major contracts (e.g., deployment partnerships) for consistency with this policy; (4) overseeing the implementation of this policy, including the allocation of sufficient resources; (5) receiving and addressing reports of potential instances of noncompliance; and (6) making judgment calls on policy interpretation and application.” For Risk Reports: “The CEO and RSO will make the ultimate determination regarding the adequacy of the risk assessment and any downstream deployment or development plans.” After approval, “the CEO and RSO will promptly share their decision(s), the underlying Risk Report, and internal feedback with both the Board and the LTBT” (Anthropic, RSP v3.0; v3.4, effective July 8, 2026). Where a decision leans on a marginal-risk argument, “explicit approval of the Risk Report by the Board and LTBT (rather than just the CEO and RSO) will be required” (v3.4).

• The seat’s checks, as the policy states them. “Changes to the RSP will be proposed by the CEO and RSO, and approved by the Board in consultation with the LTBT.” “We will regularly brief the LTBT on plans and developments related to our Risk Reports.” Noncompliance reports go to “the RSO, and at least one executive who does not report to the RSO,” and “where a report concerns the conduct of the RSO, at least one recipient will be a member of the Board.” The policy also commits that Anthropic “will not impose contractual non-disparagement obligations on employees, candidates, or former employees in a way that could impede or discourage them from publicly raising safety concerns about Anthropic” (Anthropic, RSP v3.0; v3.4).

• The policy’s history under him. Version 2.0 took effect October 15, 2024, the day of his appointment; versions 2.1 (March 31, 2025), 2.2 (May 14, 2025), 3.0 (February 24, 2026), 3.1 (April 2), 3.2 (April 29), 3.3 (May 26) and 3.4 (July 8, 2026) followed (Anthropic, RSP updates). The July 8, 2026 update “revises our threshold for automated R&D to better track the threat model of concern” and revises the internal-sharing rule to “at least 200 Anthropic employees” (Anthropic, RSP updates). On August 14, 2026 Anthropic shared its August 2026 Risk Report (Anthropic, RSP updates).

• February 24, 2026: he explains the rewrite to TIME. TIME reported that Anthropic was “dropping the central pledge of its flagship safety policy.” Kaplan: “We felt that it wouldn’t actually help anyone for us to stop training AI models,” and “We didn’t really feel, with the rapid advance of AI, that it made sense for us to make unilateral commitments … if competitors are blazing ahead.” He rejected the framing of a retreat: “I don’t think we’re making any kind of U-turn.” TIME also reports that “Executives also hoped the approach might eventually serve as a blueprint for binding national regulations or even international treaties, Kaplan claims. But those regulations never materialized.” And: “If all of our competitors are transparently doing the right thing when it comes to catastrophic risk, we are committed to doing as well or better” (TIME, Billy Perrigo, February 24, 2026). The policy text for the “Anthropic in the lead” case: “We will delay AI development and deployment as needed to achieve this, until and unless we no longer believe we have a significant lead” (Anthropic, RSP v3.0).

• December 2, 2025: his account of the next decision. The Guardian reported that “Humanity will have to decide by 2030 whether to take the” “ultimate risk” of letting artificial intelligence systems train themselves to become more powerful. Kaplan: recursive self-improvement “is in some ways the ultimate risk, because it’s kind of like letting AI kind of go.” “That’s the thing that we view as maybe the biggest decision or scariest thing to do.” He named two risks: “One is do you lose control over it?” and the security risk that such systems “fall into the wrong hands”; “I think preventing power grabs … preventing misuse of the technology, is also very important.” The article adds that he “urged international governments and society to engage” in what he called “the biggest decision” and that AI systems would be capable of doing “most white-collar work” in two to three years (The Guardian).

• Claude Character and the constitution. Anthropic’s constitution page: “Jared Kaplan worked with Amanda to create the Claude Character project in 2023, to set the direction for the new constitution, and to think through how Claude would learn to adhere to it. He also gave feedback on revisions and priorities for the document itself.” It lists him with Chris Olah and Holden Karnofsky among those who “made significant contributions to its content and development,” and the PDF author line reads “Amanda Askell,* Joe Carlsmith,* Chris Olah, Jared Kaplan, Holden Karnofsky” (Anthropic, Claude’s constitution; Anthropic PDF). See Amanda Askell, Chris Olah, Joe Carlsmith and Holden Karnofsky.

• The papers. “Scaling Laws for Neural Language Models” (submitted January 23, 2020; Kaplan, McCandlish, Henighan, Brown, Chess, Child, Gray, Radford, Wu, Amodei) reports that “The loss scales as a power-law with model size, dataset size, and the amount of compute used for training” (arXiv:2001.08361). “Constitutional AI: Harmlessness from AI Feedback” (submitted December 15, 2022) lists 51 authors, with Yuntao Bai first and Jared Kaplan last (arXiv:2212.08073). His name is also on the December 2021 HHH paper listed on the Amanda Askell profile (arXiv:2112.00861).

• Who reports to him. Fortune reported that Holden Karnofsky joined in January 2025 as “a member of the technical staff” reporting to chief science officer Jared Kaplan, working on the responsible scaling policy, per an Anthropic spokesperson (Fortune, February 13, 2025).

• The money, as filed or stated. Anthropic’s post of October 18, 2026: “When Anthropic was founded, our co-founders pledged to donate 80% of their equity holdings” (Anthropic, Investing in the public benefit); the post gives no figure for any one co-founder. The Guardian calls him “the chief scientist and co-owner” of the company and says he “has gone from being a theoretical physicist scientist to an AI billionaire” (The Guardian); that is the paper’s description, not a filing or his own statement. FEC records list “KAPLAN, JARED DANIEL,” employer ANTHROPIC, occupation CO-FOUNDER, San Francisco, with five contributions of $416 each, April 30 through August 31, 2026, $2,080 in all, to the Anthropic PBC Political Action Committee (C00946111), a corporate PAC registered April 3, 2026 (FEC, receipts; FEC, committee C00946111). No SEC filing stating his shareholding was found.

Social Persona / Impression Management

Immediate impression: The physicist at the whiteboard. His interview prose reasons from curves and thought experiments (“If you imagine you create this process where you have an AI that is smarter than you, or about as smart as you, it’s [then] making an AI that’s much smarter”) and then states the stakes in plain words (“It sounds like a kind of scary process. You don’t know where you end up”) (The Guardian).

Energy: Press-facing, not feed-facing. The two long records of his own words are interviews: TIME’s exclusive on the policy change and the Guardian’s series piece (TIME; The Guardian). No personal blog, newsletter or social account was found.

Impression management strategy: Pragmatism over principle, stated as continuity. “I don’t think we’re making any kind of U-turn” is the frame for removing a pledge; “we are committed to doing as well or better” is the replacement commitment; and the argument is that stopping would not help because competitors would go on (TIME). The strategy is open about its premise: the company’s founding premise, TIME writes, was that “to do proper AI safety research, they had to build models at the frontier of capability.” What the record adds is the seat: he is the interviewee explaining the change and the officer the changed policy names as approver.

• Social accounts and feeds. No X, Bluesky, Mastodon, newsletter or personal-site account found on his Johns Hopkins page, the Company page or in search; the Johns Hopkins page lists a curriculum vitae link and an institutional email only (JHU Physics & Astronomy). Feed: none for him; the policy’s page anthropic.com/rsp-updates is the place version changes and Risk Reports are announced and has no RSS feed linked.

Forensic Archetype Comparison

Psychometric Assessment

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

Big Five (OCEAN):

Dark Triad (held low and evidence-bound; the score measures structural position, not character):

MBTI: INTJ (“The Architect”) — reasons from a curve to a policy, builds a rule set with numbered duties, and defends a change by its premises. Rationale is the public record, not a typing session (TIME).

Threat Assessment

Score move: none (new profile). Calibrated against peers on the same 8/7/8 triple: Nicholas Carlini (hazard 67, research seat with no release approval), Ben Buchanan (74), David Sacks (74) and Palmer Luckey (74), all advisory, policy or procurement seats, and Peter Mattson (76, president of a benchmark consortium). Subject holds a named approval gate over a frontier model’s release, which puts him with Mattson above the advisory seats and well above Carlini. Against Dario Amodei (7/7/8, 71.8), who shares the final determination, the RSO seat scores a point higher on leverage because the policy names it as approver of deployment decisions; against Jack Clark (8/8/8, 82) he scores a point lower on insulation because the CEO co-decides and the Board and Trust receive each decision.

Score basis. Atk 8 is leverage: the RSO’s duties include “approving relevant model development or deployment decisions based on our risk assessments” and, with the CEO, the RSO makes “the ultimate determination” on Risk Reports; Anthropic named him to the seat (Anthropic, RSP v3.0; Anthropic). It is not a 9 because the decision is joint with the CEO, and no page after October 2024 names the current holder. Def 7 is insulation: co-founder, with changes to the policy “approved by the Board in consultation with the LTBT” (Anthropic, RSP v3.4); it is not an 8 because the same text routes noncompliance reports about the RSO to a Board member. Hp 8 is durability: a 2020 paper, a 2022 method and a policy at version 3.4 (arXiv:2001.08361; arXiv:2212.08073; Anthropic, RSP updates); it is not a 9 because the policy is voluntary and was rewritten once already.

Alliance Map

The overlap is the finding. The Corporate seat that writes the lab’s scaling bets, the Corporate seat that signs its release decisions, and the pledge that gives most of the equity away if kept run at the same time, and they converge on a company whose growth the first creates, the second licenses and the third is meant to redeem. None of it is concealed; the policy states the seat’s powers in public.

Alignment Analysis

Stated alignment: “If all of our competitors are transparently doing the right thing when it comes to catastrophic risk, we are committed to doing as well or better” (TIME); a decision on self-improvement for “international governments and society” to weigh (The Guardian).

Observed alignment: Holds the seat that approves release decisions under the lab’s own policy; co-proposes the policy’s changes; explained to TIME why the unilateral commitment was removed; and tells the Guardian that the self-improvement decision belongs to governments and society. The policy that replaced the pledge keeps conditional delay: “We will delay AI development and deployment as needed to achieve this, until and unless we no longer believe we have a significant lead” (Anthropic, RSP v3.0).

Gap assessment: No documented gap between what he says and what he does. The structural item is that three functions meet in one seat: the officer who applies the rule (“approving relevant model development or deployment decisions”), the officer who proposes its amendment (“Changes to the RSP will be proposed by the CEO and RSO”), and the officer to whom noncompliance reports go (“the RSO, and at least one executive who does not report to the RSO”). The Board approves changes and the Trust is briefed; both are named in the policy, and neither is described as able to veto a release decision. The record does not settle whether the seat’s checks work, and for the officer it never needs to.

Interest: (the faction’s stake in each position; a documented fact about the position, never a claim about motive.)

1. Position: do not make unilateral commitments to stop while competitors proceed. “We didn’t really feel, with the rapid advance of AI, that it made sense for us to make unilateral commitments … if competitors are blazing ahead” (TIME).
   • What adoption would pay the subject’s firm. The old pledge barred training above a capability level without adequate safeguards; the new policy conditions delay on being in the lead. TIME reports the company “raised $30 billion in new investments, valuing it at some $380 billion” in the same month, with annualized revenue growing at “10x per year.” The documented fact is the timing and the scale of what continued training is attached to; no source says the figures drove the decision.
   • His own best case, steelmanned. TIME quotes the new policy’s introduction: “If one AI developer paused development to implement safety measures while others moved forward training and deploying AI systems without strong mitigations, that could result in a world that is less safe. The developers with the weakest protections would set the pace, and responsible developers would lose their ability to do safety research.” Kaplan: stopping would “most likely lose relevance as an innovator who understands the frontier of the technology, in a scenario where others are going ahead and we’re not actually contributing any additional risk to the ecosystem.”
   • Who else gains from the same rule. Every frontier lab whose commitments are conditional: TIME notes that “No rivals made quite as overt a promise to pause AI development.” Investors in any lab whose training run is not halted by its own pledge.
2. Position: the self-improvement decision belongs to governments and society. Kaplan “urged international governments and society to engage” in what he called “the biggest decision” (The Guardian).
   • What adoption would pay the subject’s firm. The interview specifies no rule, so nothing is created or exempted. What is documented is the sequence: a lab asks outsiders to decide a question while its own policy keeps the internal decision with its CEO and RSO.
   • His own best case, steelmanned. “once no one’s involved in the process, you don’t really know. You can start a process and say, ‘Oh, it’s going very well’… But you don’t know — it’s a dynamic process.”
   • Who else gains from the same position. Governments and international bodies asked to hold the authority; competing labs, which can cite the same decision as theirs to share.
3. Position: frontier policies as the blueprint for binding rules. “Executives also hoped the approach might eventually serve as a blueprint for binding national regulations or even international treaties, Kaplan claims” (TIME).
   • What adoption would pay the subject’s firm. A statute modeled on the policy would make its Risk Reports, thresholds and roadmaps the compliance form; the company already publishes each of them under the RSP (Anthropic, RSP updates). The rule would not exist yet: “those regulations never materialized” (TIME).
   • His own best case, steelmanned. Voluntary commitments cannot hold alone when “the Trump Administration has endorsed a let-it-rip attitude,” in TIME’s words, and no federal law is on the horizon; his stated hope was binding rules that bind rivals too.
   • Who else gains from the same rule. The other developers that publish frontier safety policies (METR counts twelve on its common-elements page); evaluators such as METR whose methods the policies cite.

Nothing above is a verdict that the positions are wrong. The finding is the alignment of roles: the author of the scaling curve, the officer who approves releases against it, and the interviewee who explains why the pledge to stop was removed are one person.

Convergent Drive Classification

Self-preservation: The co-founder pledge is stated for the group; no individual figure is filed or self-stated (Anthropic, Investing in the public benefit). He keeps a faculty listing at Johns Hopkins on the department page (JHU Physics & Astronomy). Goal preservation: The goal, frontier research done safely, is restated as the premise of the policy change: “to do proper AI safety research, they had to build models at the frontier of capability” (TIME). Resource acquisition: Holds the seat that allocates the policy’s resources: the RSO’s duties include “overseeing the implementation of this policy, including the allocation of sufficient resources” (Anthropic, RSP v3.0). Self-improvement: From a physics chair to a lab’s science chief to an officer’s seat, with the policy at nine published versions in three years (JHU Physics & Astronomy; Anthropic, RSP updates).

Subject is not an AI system. The drives appear anyway — in the chief scientist who signs the releases, proposes the rules the releases are measured by, and explains in public why the old rule was dropped.

Public footprint: No personal accounts found. TIME interview February 24, 2026; Guardian interview December 2, 2025; policy page anthropic.com/rsp-updates; department page JHU Physics & Astronomy.

Sources: Anthropic — Announcing our updated Responsible Scaling Policy, Oct 15, 2024; Anthropic — Responsible Scaling Policy v3.0; Anthropic — Responsible Scaling Policy v3.4; Anthropic — RSP updates; Anthropic — Company; Anthropic — Claude’s constitution; Anthropic — Investing in the public benefit, Oct 18, 2026; TIME — Anthropic Drops Flagship Safety Pledge, Feb 24, 2026; The Guardian — ‘The biggest decision yet’: Jared Kaplan on allowing AI to train itself, Dec 2, 2025; Fortune — Anthropic hires Holden Karnofsky, Feb 13, 2025; arXiv:2001.08361; arXiv:2212.08073; JHU Physics & Astronomy — Jared Kaplan; FEC — Anthropic PBC Political Action Committee.