CHARLES TEAGUE

Behavioral Archetype

THE KEEPER OF THE RULER — Subject is the chief executive of the nonprofit that maintains the software other people measure frontier models with. Meridian Labs writes no ranking and publishes no score about a company; it keeps Inspect, the harness the UK AI Security Institute built and open-sourced in May 2024, and since 7 May 2026 it keeps Petri, the alignment auditor Anthropic donated. J.J. Allaire is the name on the most commits. Subject is the name on the incorporation papers, the Boston address and the CEO line, and the second name on the commit list of the repository the institutes run (GitHub, inspect_ai contributors). The hazard is the seat of the person who decides who maintains the tool, because every result run in the tool inherits what the tool computes, and the steward’s own pages name nobody who pays for it. This profile scores the seat, not the man.

Essence Indicators

• Chief Executive Officer and co-founder, Meridian Labs. Meridian’s team page lists him as “CEO, Co-founder”: “Charles co-founded Meridian Labs in 2025 to advance the public understanding of AI systems through comprehensive evaluation tools and research.” The IRS record for the organization, Meridian Research Labs Inc, prints its address as “% CHARLES TEAGUE,” 24 School Street, Boston, with a 501(c)(3) ruling dated March 2025 (Meridian Labs, Team; ProPublica Nonprofit Explorer). The Careers page’s operations role is one that will “work directly with our CEO” (Meridian Labs, Careers).

• What is documented about the money. Meridian’s home, About, Team, Careers and Projects pages name no funder, donor, grant or contract value. ProPublica’s page for the organization carries no extracted financial data (no Form 990 digitized). The Careers page prices the payroll it is hiring against: “Senior and principal levels · $175,000 – $325,000” for engineering and research staff, with the operations role described as one for “a small technical non-profit” (Meridian Labs, Careers). The two documented transfers into the organization are in-kind: Anthropic “handed over” Petri’s development to it on 7 May 2026, and the UK institute’s engineers work beside it on Inspect (Anthropic, Donating our open-source alignment tool; Meridian Labs, Introducing Petri 3.0). The only named money near him is RAND’s, below.

• Inspect, in his own organization’s two wordings. Meridian’s About page: “Our flagship project is Inspect AI, an evaluation framework built by Meridian’s founding team in collaboration with the UK AISI.” Its home page, under the same project: “Framework for systematic evaluation of large language models, built by the UK AISI in collaboration with Meridian.” The Inspect documentation: “Inspect is a framework for frontier AI evaluations developed by the UK AI Security Institute and Meridian Labs.” The UK institute’s playbook: “In May 2024, the AI Security Institute open-sourced Inspect AI, a toolkit for evaluating the capabilities of frontier AI models” (Meridian Labs, About; Meridian Labs; Inspect; UK AISI, Releasing AISI’s Engineering Playbook). The same playbook attributes a quotation to “JJ Allaire, creator of Inspect and founder at Meridian Labs,” and names no other Meridian person. His bio gives his own part: before Meridian he “was a Technology and Security Policy Fellow at the RAND Corporation, where he helped develop Inspect AI” (Meridian Labs, Team).

• His part in the repository. GitHub’s contributor list for UKGovernmentBEIS/inspect_ai shows the account dragonstyle, named “Charles Teague” on its profile, second with 1,142 contributions, behind jjallaire (2,852) and ahead of ransomr (927), jjallaire-aisi (336) and epatey (311); the first commit credited to that account is dated 15 July 2024 (GitHub, inspect_ai; GitHub, dragonstyle). Meridian’s team page links that GitHub account from his entry (Meridian Labs, Team). Four of the top five accounts (jjallaire, dragonstyle, ransomr, epatey) are the GitHub handles Meridian’s team page links for Allaire, Teague, Ransom Richardson and Eric Patey.

• Where Inspect runs. Meridian’s About page: “Inspect AI is now the standard framework for government organizations including UK AISI, US CAISI, EU AI Office, Japan AISI, Korea AISI, and others. It is also the standard framework for research organizations including METR, Apollo, Epoch, SecureBio, Redwood, and RAND.” The independent record covers part of that list. The UK institute’s playbook: “METR migrated from their in-house Vivaria platform to Inspect”; “Apollo Research deprecated their internal evaluation framework after a three-month trial of Inspect”; “Other government institutions have adopted Inspect and the sandboxing toolkit for their own evaluation work”; and the repository has “240+ contributors to inspect_ai alone.” The US Center for AI Standards and Innovation’s evaluation of DeepSeek models records “modifying the benchmark to run in the Inspect evaluation framework.” METR’s Vivaria repository says “METR is transitioning its internal tooling from Vivaria to Inspect” (Meridian Labs, About; UK AISI; NIST CAISI; GitHub, METR/vivaria). The EU, Japanese and Korean uses rest on Meridian’s statement alone. See METR, Apollo Research, Epoch AI and SecureBio.

• RAND, twice, with a changed title. RAND’s November 2025 report on the biological knowledge of frontier models lists him among 23 authors and describes him: “Charles Teague is a Center for AI, Security, and Technology fellow at RAND. He conducts technical research on frameworks and methodologies for evaluating large language model capabilities.” Its method line: “All benchmarks were implemented in the United Kingdom AI Security Institute’s Inspect framework to enable standardized, repeatable execution.” RAND’s January 2026 report on LLM agents and benchtop DNA acquisition lists him among 14 authors and now reads “Charles Teague is an adjunct at RAND,” and states “Our core evaluation was implemented using the Inspect AI library, a framework for LLM evaluations developed by the UK AI Security Institute” (RAND RR-A3797-1; RAND RR-A4591-1). RAND’s funding statements: the first report “was completed through funding provided by DALHAP Investments Ltd., Effektiv Spenden, Ergo Impact, Founders Pledge, Charlottes och Fredriks Stiftelse, Good Ventures, Jaan Tallinn, Longview, Open Philanthropy, and Waking Up Foundation”; the second “was made possible thanks to generous contributions from Chris Anderson and Jacqueline Novogratz, Coefficient Giving, High Tide Foundation, The Pew Charitable Trusts, Sea Grape Foundation, Jaan Tallinn, and Valhalla Foundation as part of The Audacious Project,” with “RAND clients, donors, and grantors have no influence over research findings or recommendations” (RAND RR-A3797-1; RAND RR-A4591-1). See RAND, Open Philanthropy, Good Ventures, Jaan Tallinn and Longview Philanthropy.

• Before Meridian. His team-page entry: “Previously, Charles worked at Posit, PBC, where he created the Quarto scientific publishing system. Charles also founded Lose It! co-founded Onfolio, and was a VP of Engineering at Allaire Corp.” Quarto’s own About page lists “Charles Teague ( @dragonstyle )” among the core team and says Quarto’s development “is sponsored by Posit, PBC.” Lose It!’s page says the app “was created” in 2008 and does not name him (Meridian Labs, Team; Quarto, About; Lose It!, About). His blog, titled “The Tech Blog of Charles Teague,” carries 26 posts, the most recent dated 26 January 2024, on Quarto releases and extensions (Dragonstyle). See J.J. Allaire, who founded Posit.

• AEF-1 author and letter signatory. The AEF-1 standard’s 26-author list, which the AI Evaluator Forum publishes, includes “Charles Teague,” affiliation Meridian Labs, beside Beth Barnes, Christopher Painter (Chris Painter), Miles Brundage and Jaime Sevilla. The Forum’s letter of 18 September 2026 lists “Charles Teague CEO, Meridian Labs” among its notable signatories, and says “All signatories sign in their personal capacity. Affiliations are listed for identification only.” The letter’s conditions include that frontier companies “should rely on evaluators that are meaningfully independent, that maintain full editorial control, and that disclose and mitigate potential conflicts of interest” (AI Evaluator Forum, AEF-1 PDF; AI Evaluator Forum, letter).

• Petri changes hands (7 May 2026). Meridian’s post “Introducing Petri 3.0” lists seven authors, with “Charles Teague 1” last, “1 Meridian Labs”; his co-authors Allaire and Fronsdal also carry “UK AISI Research Affiliate,” and three others are UK AI Security Institute staff. The post: “We took on Petri’s development because this is exactly the kind of work we exist to support.” “Our goal as stewards is simple: keep Petri useful for the projects already building on it, and make it more hackable for the ones that haven’t started yet.” “Anthropic will continue to support Petri and use it in its own alignment assessments.” It adds that the UK institute “built its alignment evaluation pipeline on Petri to test frontier models for research-sabotage propensity.” Anthropic’s post of the same day: “We have handed over its development to Meridian Labs, an AI evaluation nonprofit” (Meridian Labs, Introducing Petri 3.0; Anthropic). Meridian’s Careers page adds the other direction: “Petri alignment audits by our team appear in recent Anthropic model cards” (Meridian Labs, Careers). See Kai Fronsdal and Alexandra Abbas.

Social Persona / Impression Management

Immediate impression: The committer with a title. The public record is a three-sentence team-page bio, a GitHub account with the commit history, a Quarto blog that stopped in January 2024, two RAND bios and a name in author lists. No talk, interview, podcast or testimony under his name was found, and he is last of seven on the one Meridian post that carries his byline.

Energy: Low-profile and artifact-centered. His output is code, release notes and co-authored documents. Of Meridian’s seven blog posts, Allaire and Abbas wrote three each (Inspect Viz, Scout and long-horizon agents; Harbor, Flow and cloud sandboxes) and the one bearing Teague’s name is the Petri handover (Meridian Labs, Blog).

Impression management strategy: Provenance by wording. Meridian describes Inspect two ways on two pages (founders “in collaboration with” the UK institute; the institute “in collaboration with Meridian”), the docs credit both, and the institute’s playbook says the institute open-sourced it. Each describes the same arrangement from a different side, and a reader who stops at one page gets a different founder. The exhibit that carries the most weight is the About sentence that the framework is “now the standard framework” at five governments’ institutes and six research organizations, an adoption claim the three independent pages above support in part.

• Social accounts and feeds. No personal X account found. GitHub dragonstyle, linked from his entry on Meridian’s team page; personal blog charlesteague.com (“Dragonstyle: The Tech Blog of Charles Teague”), whose advertised feed address returns no feed; no newsletter found. Meridian’s own X account is @meridianlabs_ai and its blog feed is meridianlabs.ai/blog/index.xml, both linked from meridianlabs.ai; the one post under his name appears there.

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: ISTJ (“The Logistician”) — a maintainer whose public footprint is commit history and release process, who appears in print mostly through author lists. Rationale is the public record, not a typing session (GitHub).

Threat Assessment

Score move: none (new profile). Calibrated against Maxime Labonne (7/5/7, hazard 62, the one peer on identical values: head of post-training at a model company and author of widely used open-source tooling, so the same leverage over what others build on and no outside funder record to weigh), J.J. Allaire (7/6/8, hazard 62, co-founder and top committer of the same repository, with a public benefit corporation’s record behind his other seat), Marius Hobbhahn (7/7/7, hazard 60, runs an evaluator that moved onto Inspect and publishes its funding) and Conrad Stosz (5/5/5, hazard 40, a governance seat with no repository). Subject sits one point below Allaire on insulation and durability (second committer, not first; no earlier institution of the Posit kind), and one point below Labonne on hazard because his organization publishes no scores of its own.

Score basis. Atk 7 is leverage: Meridian’s About page says Inspect “is now the standard framework” at the UK, US, EU, Japanese and Korean institutes and at METR, Apollo, Epoch, SecureBio, Redwood and RAND; CAISI’s DeepSeek evaluation, the UK institute’s playbook and METR’s repository support use at three of them (Meridian Labs, About; NIST CAISI; UK AISI; GitHub, METR/vivaria); it is not an 8 because no statute or procurement standard names Inspect and the institute “open-sourced” it, so adoption is voluntary. Def 5 is insulation: no Meridian page names a funder, a budget or a donor and the IRS record carries no financial data (ProPublica Nonprofit Explorer); the AEF-1 standard he co-authored lists “a meaningful fraction of the evaluator’s funding (either revenue or donations) comes from a system provider, its employees, or its direct competitors” as a disclosure category (AEF-1 PDF); it is not a 6 because the number cannot be checked against a published budget. Hp 7 is durability: the repository is MIT licensed with “240+ contributors” per the UK institute, METR and Apollo moved their internal frameworks onto it, and Quarto is a shipped system with his name on its core-team list (UK AISI; Quarto); it is not an 8 because the institute that created Inspect holds the repository and Meridian is a 2025 organization.

Alliance Map

The overlap is the finding. The person who keeps the harness is at the same time the CEO of a nonprofit whose funders are unnamed, an adjunct at a research center whose funders are named and philanthropic, a co-author of the standard that asks evaluators to disclose funding, and a signatory of the letter that asks frontier companies to embed evaluators, while his organization maintains a lab-donated auditor and runs audits that appear in that lab’s model cards. Each tie is printed on its own page; none is hidden. What no page prints is who pays Meridian.

Alignment Analysis

Stated alignment: Meridian’s mission, which he leads: “to advance AI safety, alignment, and security work by creating a common platform shared by governments, non-profits, academia, and model developers” (Meridian Labs, About); and the letter he signed: “We believe that all frontier AI companies should embed evaluators to independently assess AI risks,” with evaluators that “disclose and mitigate potential conflicts of interest” (AI Evaluator Forum, letter).

Observed alignment: Writes and maintains the common platform (1,142 contributions) and takes custody of a lab’s alignment auditor; the UK institute’s engineers are co-authors and “built its alignment evaluation pipeline on Petri”; Meridian’s own team runs Petri audits that “appear in recent Anthropic model cards” (GitHub; Meridian Labs, Introducing Petri 3.0; Meridian Labs, Careers).

Gap assessment: No documented gap between what he writes and what he does. The structural item is the one every harness shares: the tool, the tester and the grader sit close together, and for Inspect and Petri the steward’s funding is the part that no page states. The letter asks for disclosure of conflicts; Meridian’s pages disclose none and name none, and no page says whether any exist. The record cannot settle whether the steward’s priorities and its users’ could diverge, and for a maintainer of public code with an open license it does not have to: a user can read what a score computes.

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

1. Position: a common, open-source evaluation platform shared by governments, nonprofits, academia and model developers (Meridian’s About page, which he leads; Meridian Labs, About).
   • What adoption would pay Meridian. Every institute or lab that standardizes on Inspect, Scout, Flow or Petri depends on the party that maintains them; the About page lists “Conducting pre-deployment testing using our tools” as part of Meridian’s work. No fee, contract or grant value is published.
   • His organization’s best case, steelmanned. “Providing the infrastructure for AI Safety Institutes, research organizations, and lab safety teams to transfer their work into broadly-available open source projects”; the code is open and MIT licensed; and the alternative to a common harness is each evaluator maintaining its own, as METR and Apollo did before moving.
   • Who else gains from the same rule. The UK institute, which offloads upkeep of a tool it created while its researchers remain Meridian co-authors; METR and Apollo, which retired in-house frameworks; the lab that donated Petri, whose own model cards report runs of it (UK AISI; Anthropic).
2. Position: frontier companies should embed independent third-party evaluators with employee-like access (the September 2026 letter he signed; AI Evaluator Forum, letter).
   • What adoption would pay Meridian. Embedded and third-party assessment is work that Meridian already lists among its own (“Conducting pre-deployment testing using our tools”) and does for Anthropic’s model cards; a rule that makes more companies host evaluators widens the market for that work. The letter’s own conditions (independence, editorial control, conflict disclosure) also describe what Meridian would have to publish to qualify, and its pages publish none of it.
   • The letter’s best case, steelmanned. “To be credible, embedded third-party evaluations must have scientific objectivity, transparency, independence, and robust protections against interference from the evaluated companies”; the letter calls the embedded model “a complement to, rather than a replacement for, broader efforts.”
   • Who else gains from the same rule. The evaluators who wrote AEF-1 with him (METR, Transluce, AVERI, SecureBio, Epoch AI, GovAI authors, the Collective Intelligence Project); GovAI’s Markus Anderljung, who signed the same letter and whose GovAI published “Embedded Assessments for Frontier AI,” on which he is an author, three days after it; and the frontier developers the letter says “call for embedding third-party organizations” (AI Evaluator Forum, AEF-1; GovAI).

Nothing above is a verdict that the positions are wrong. The finding is the alignment of interests: the person who signs a letter asking evaluators to disclose conflicts runs an evaluator that names no funder, and the market that letter would create is the one his organization already works in.

Convergent Drive Classification

Self-preservation: The steward role is held through the repository’s own history: second-most commits, the CEO line, the Boston filing, and a lab’s donation of a second tool in May 2026 (GitHub; Anthropic). Goal preservation: The goal, a common open platform for testing frontier models, is restated on the About page, the Careers page and the Petri post (Meridian Labs, About; Meridian Labs, Careers). Resource acquisition: Takes custody of tools other institutions would rather not own (Inspect’s upkeep, Petri, Bloom) and hires at $175,000 to $325,000 senior and principal bands with no funder named (Meridian Labs, Careers). Self-improvement: From a publishing-software founder to Quarto’s core team to a RAND fellowship to the CEO of an evaluation nonprofit (Meridian Labs, Team; RAND RR-A3797-1).

Subject is not an AI system. The drives appear anyway — in the maintainer of the harness that grades the models, at a steward that publishes its code and not its funders.

Public footprint: GitHub dragonstyle (linked from Meridian’s team page); blog charlesteague.com; Meridian X @meridianlabs_ai and blog feed meridianlabs.ai/blog/index.xml. RAND reports: biological knowledge benchmarks, benchtop DNA acquisition.

Sources: Meridian Labs, Team; Meridian Labs, About; Meridian Labs, Careers; Meridian Labs; Meridian Labs, Introducing Petri 3.0; Meridian Labs, Blog; Anthropic, Donating our open-source alignment tool; Inspect; UK AISI, Releasing AISI’s Engineering Playbook; UK AISI, Announcing Inspect Evals; GitHub, UKGovernmentBEIS/inspect_ai; GitHub, dragonstyle; GitHub, METR/vivaria; NIST CAISI, Evaluation of DeepSeek AI Models; RAND RR-A3797-1; RAND RR-A4591-1; AI Evaluator Forum, letter; AI Evaluator Forum, AEF-1 PDF; Princeton HAL, creators; Quarto, About; Lose It!, About; Dragonstyle, charlesteague.com; ProPublica Nonprofit Explorer, Meridian Research Labs Inc.