← The Independent 100: a study

davidad

#12 in The Frontier Outside, the curator’s rank and section @davidad Relevance to our work: 3 of 5

Bracketed tags are evidence labels: [Verified] means we read the page, file or API response ourselves, [Reported] means a source says so and we did not check the underlying thing, and [Inference] is our own reasoning.

Credit

The Independent 100 list was curated by dan (@irl_danB) and published on 2026-09-24 at independent.prose.md (announcement). The ranks and section names are his, and number 100 is his open wild-card slot. This study is ours and is not endorsed by him or by anyone on the list.

What they publicly do

Technical Advisor, formerly Programme Director, of ARIA's £59m Safeguarded AI programme, which now aims to build an open-source mathematical assurance toolkit that lets fleets of AI agents produce formally verified artifacts, applied first to cybersecurity [Verified aria.org.uk/opportunity-spaces/trust-everything-everywhere/safeguarded-ai]. Posts on X about alignment, RLVR reward design and formal verification [Verified x.com/davidad/status/2098870677382021294]; started his own alignment initiative in April 2026 [Reported centerforhumanetechnology.substack.com/p/have-we-trained-ai-to-lie-to-itself].

Public work

  • Safeguarded AI programme thesis v2 (ARIA, May 2026) paper no license stated last activity 2026-05

    Programme he launched: TA1 'Tooling' (open-source assurance toolkit for agent fleets) and TA2 cybersecurity applications with eight funded projects [Verified ARIA page; PDF not read; authorship of v2 not attributed to him].

  • geolog-zeta repo license file present, not recognised by GitHub last activity 2026-03-20 · 6 GitHub stars on 2026-09-24

    Self-described vibe-coded prototype of a version-controlled database modeled on geometric logic, in Rust [Verified repo].

  • chit repo no license file last activity 2023-07-23 · 19 GitHub stars on 2026-09-24

    Early experiments toward a version-control system for structured data, in Rust [Verified repo].

  • Compute thresholds: proposed rules to mitigate risk of a lab leak (LessWrong, 2023) blog no license stated last activity 2023

    Rules for large training runs including a recursive provenance audit trail; he re-cited it on 2026-09-12 [Verified x.com/davidad/status/2098853192540778760; post body not read].

  • Towards provably safe AI (Gradient podcast) podcast no license stated last activity 2024-09

    Interview on his provable-safety approach [Reported via ARIA page link].

Licenses of GitHub repositories are from the authenticated GitHub license API on 2026-09-25, one row per repository in evidence/licenses.jsonl (“no license file” means the API found none; “license file present, not recognised by GitHub” means GitHub could not classify it). Other licenses are as the source states. Check the license file before reusing anything.

Relevance to our work

3 of 5 lenses: verification, research-method, security

The programme's thesis (AI agents producing machine-checked artifacts because verification has not kept pace with agent-written code) is the same problem franken-research's rigor practices address [Inference]. His posts on grading rollouts by contrastive partial ordering rather than scalar RLVR bear on how LLM judges and graders are designed [Inference]. Little adoptable code from him directly.

Scored 0 to 5 against the study’s relevance rubric: 5 means public, openly licensed work squarely in our lanes that we could study or adopt now, and 1 means little connection to our work. The score is one reviewer’s judgement of how much this public work bears on our own work. It is not a ranking of people and says nothing about the person or the quality of their work.

What we would study next

  1. Safeguarded AI programme thesis v2 (aria.org.uk/media/ikrkutfk/safeguarded-ai-programme-thesis-v2.pdf). Defines what an 'assurance toolkit for agent fleets' must provide; a checklist for agent verification practice. Effort: medium.
  2. ARIA TA2 cybersecurity funded projects (aria.org.uk/opportunity-spaces/trust-everything-everywhere/safeguarded-ai/funded-projects?tabId=applications). Includes work on containing AI agents; possible sources on prompt-injection defence and sandboxing. Effort: small.
  3. His RLVR / partial-ordering posts (x.com/davidad/status/2098870677382021294). Short argument for comparing rollouts pairwise instead of scalar rewards; relevant to judge and grading design. Effort: small.

Before reusing anything: geolog-zeta: LICENSE text offers MIT or Apache-2.0 but GitHub's API reports NOASSERTION; it is self-described as a vibe-coded prototype. chit has no license file (default all rights reserved). ARIA TA1 toolkit code availability not checked.

Sources

Overall confidence in this record: Medium. Sources were read on 2026-09-24 and every link was checked on 2026-09-25. The full record is line 12 of people.jsonl.

Limits

Not verified: davidad.org is JS-rendered and returned only a stub. The April 2026 'own alignment initiative' has only a secondary source. Whether any Safeguarded AI toolkit code is public yet. No primary text was found for the 'Wisdom Basin hypothesis' he mentions.

This record comes from a first-pass survey of public material by an AI agent session on 2026-09-24, edited for publication on 2026-09-25. Nothing was installed or run for it. Roles, stars and licenses change; see the study’s limits.

Request a correction

Is something about this record wrong or out of date, or would you rather not be covered? Open a correction (a GitHub issue form with this page’s address filled in). A request to remove a record is honoured without argument; accepted corrections are recorded in the changelog with the date.

Help fix it

Each opens a GitHub issue form with this page’s address filled in. You need a GitHub account.

Made by Joshua Nowak (ZestStream) with AI coding agents. Not affiliated with or funded by Jeffrey Emanuel; Joshua uses his public tools and is a paying subscriber to jeffreys-skills.md.

Source on GitHub MIT licensed