Observed 2026-09-09. The required whole-case review in forest-reviews.md was read before this pass. Only this file was written; no canonical case changes. Research stops here pending the root's whole-case reassessment.
The documentary mechanism is a jointly considered research competition with differentiated award conditions. It gives the supporting parties a way to choose research and attach conditions to their awards. It does not establish regulatory authority, a deployed admission gate, paid partner shares, or a common body controlling all AI agents. The clearest new instrument is the call-specific FAQ's optional Google.org funding route; public announcement wording alone conceals that distinction.
Amount, parties, administration and timing
Google DeepMind's own announcement, dated 11 June 2026, describes a worldwide technical-research fund of up to US$10 million alongside Schmidt Sciences, ARIA and CAIF, supported by Google.org. It expects awardees in autumn 2026. This is an advertised ceiling, not evidence of an executed US$10 million grant or money received. The page discloses neither partner amounts nor payment dates. [DeepMind announcement](http[local research file]
CAIF's announcement likewise uses the ceiling. Its article displays 11 June 2026, while the site's listing showed 16 July 2026; retain the conflict rather than invent a second launch. CAIF's foundation page marked applications closed when inspected. Its footer identifies charity 1201294 and company 13485176. Neither page identifies the joint fund as a separate legal entity. [CAIF announcement](http[local research file] [foundation page](http[local research file]
The Schmidt-branded application portal and Schmidt contact establish an administrative front door. They do not establish Schmidt as sole grantor. Named brands, particularly Google.org and Google DeepMind, should not be silently substituted for an award's contracting legal entity. No executed award or contribution agreement was acquired. [Actual application portal](http[local research file]
Public call and selection
The portal offers Tier 1 awards up to US$300,000 and Tier 2 awards of US$300,000–1 million for 1–2 years. Its selection criteria cover agenda fit, scientific rigour, impact, philanthropic fit, feasibility, expertise and cost. Joint supporting parties consider applications; selected funders may add limitations or guidance, agreed in final award documents. Lobbying, legislative influence and political activity are excluded. Schmidt-funded indirect costs are capped at 10%.
The portal table gives 9 August 2026, 11:59pm AoE, consistent with its UI's 10 August, 07:59am EDT. The announcements instead say 8 August AoE. No acquired change history explains the one-day difference. Opening is shown as 11 June. This preserves observed versions without asserting why the deadline changed. [Portal, Timeline, Funding Levels, Eligibility and Selection Criteria](http[local research file]
Call-linked FAQ: differentiated award conditions
The FAQ is linked directly from the application portal. These are public descriptions of intended conditions, not an acquired executed agreement.
| Route / issue | Terms disclosed |
|---|---|
| Eligibility | Charitable/scientific/educational recipients; commercial participants cannot lead. Independent researchers need an eligible host or fiscal sponsor before funding. Schmidt uses NGOsource for non-US equivalency. |
| Schmidt funding | Annual gifts/grants depend on progress and reporting; subsequent years are not assured. IP normally stays with awardees unless the agreement says otherwise. |
| Google.org option | Applicants may opt out and remain eligible for Schmidt, ARIA and CAIF consideration. Google.org-funded code/inventions require Apache 2.0; other specified IP requires CC-BY 4.0, within three months. Remaining IP rights go into the public domain; patent applications are prohibited. Data must be freely searchable, subject to privacy law. Funded projects must comply with Google AI Principles. |
| Limits | Conditional renewal is not a demonstrated termination power over already-paid funds. Neither this FAQ nor the inspected call identifies reviewers, recusal rules, voting shares, publication-approval procedures or executed allocations. |
Locators: export lines 9–22 eligibility; 55–75 funding and international administration; 101–103 IP; 105–121 Google funding conditions. The FAQ includes some general Schmidt wording referring to approximately three years; the specific call says 1–2 years, which governs the duration reported here. [Call FAQ](http[local research file] [public text export inspected](http[local research file]
The linked Google principles currently address innovation, responsible development/deployment and collaboration, including human oversight, safety testing, privacy and intellectual property. The call FAQ does not identify a dated edition, change-control rule, dispute process or a party's exclusive interpretation right. Reading this webpage does not supply those missing contractual terms. [Google AI Principles, substantive principles and implementation sections](http[local research file]
Generic policies kept separate
Schmidt's general FAQ identifies Schmidt Sciences LLC, a 501(c)(3) nonprofit subsidiary of The Eric and Wendy Schmidt Fund for Strategic Innovation. The displayed EIN 46-3460261 belongs to that parent; it must not be assigned to the LLC. General overhead policy permits none without written agreement; where allowed, the normal ceiling is 10%. Thus the call's cap is not an automatic entitlement. Its general policy also discusses protected enterprise AI use during proposal processing; this is not proof that AI evaluated this competition. These are organization-wide policies, not acquired award clauses. [Schmidt general FAQ, organization, indirect-cost and AI-use questions](http[local research file]
No separate ARIA announcement or award agreement assigning its contribution was acquired in this pass. ARIA's separate Scaling Trust programme amounts must not be treated as its share of this joint call. Nor does listing Google DeepMind as a joint supporting party establish that it supplies cash separately from Google.org.
What the agenda can and cannot show
Schmidt explicitly credits DeepMind's Distributional AGI Safety work, ARIA's Scaling Trust thesis and CAIF's Multi-Agent Risks from Advanced AI report as inspirations. The agenda includes identity/authentication, admission controls, proof of agent/human credentials, platform vetting and revocation as research topics. It prioritizes strengthening existing infrastructure partly because incumbent protocols may become entrenched. These are concrete agenda choices relevant to the assurance inquiry.
Counterevidence matters: the programme excludes commercial product development, purely nontechnical policy work, and solutions that depend on one privileged principal controlling every agent. It funds research on coordination among independently controlled agents; an identity research topic is not a universal identity requirement. No acquired record selects an operational gatekeeper or compels deployment. [Schmidt research agenda, introductory inspirations, Agent Infrastructure and Out of Scope](http[local research file]
Source custody and reading scope
| Source | Actual scope inspected / state |
|---|---|
| CAIF foundation | Scoped current-call block and footer; web reader. No whole-site knowledge claimed. |
| CAIF announcement | Full 72-line web rendering, including displayed date, amounts, named parties and timetable. |
| DeepMind announcement | Full substantive announcement in 160-line web rendering; navigation not used as evidence. |
| Schmidt portal | All substantive public call text through selection criteria in a 268-line rendering; second copy of the body is duplicate page output. No private application submitted or applicant account entered. |
| Call-specific FAQ | Entire public text export read, 14,895 bytes, HTTP 200, text/plain UTF-8. Export SHA-256: aa91f1a4a06b83602f38639e5711e1d3412f1ea947f5371699fddc43d7a16c94. Public document UI yielded a JavaScript shell; ordinary public export succeeded as the single substitute. Bytes were read in memory and are not retained as an original file. No publication/revision date visible. Line locators refer to this exact export. |
| Schmidt agenda | Substantive programme introduction, agenda, research topics and exclusions. The alternate /scaling-ai-safety-for-a-multi-agent-world/ page substantially repeats the agenda; it is not an independent witness. |
| Schmidt general FAQ | Organization identity, indirect-cost and AI-use sections. Generic policy scope retained. |
| Google AI Principles | Substantive principles and implementation sections, web lines 124–169. Current linked text only; no historical incorporated version inferred. |
Web renderings are inspected source views, not locally preserved originals; no original-file hashes are invented for them. Public webinar links were identified but their recordings were not watched. Closed CAIF accounts and other prior blocked routes were not reopened.
Narrow integration and stopping point
Suitable relations are: the named parties jointly consider applications under this call; Schmidt hosts the application front door; Google.org offers a conditional funding route; applicants can decline that route without leaving the competition; the agenda attributes inspiration to three named research programmes/documents. Each belongs to the research-funding layer. Do not convert these into paid amounts, veto rights, government endorsement of findings, operational authority or donor instructions over a particular result.
The next decisive records are the joint selection/conflicts protocol, the contribution or administration agreement identifying contracting entities and committed shares, and an award-specific agreement plus disbursement record. Schmidt's programme administrator is the likely first custodian of the selection process; each actual award's grantor and recipient would hold its conditions and receipts. These are record targets only: no outreach, further branch or acquisition follows this pass before the whole-case reassessment.