01.AI

• No reference whatsoever to Indigenous data sovereignty, CARE Principles, or relational/embodied knowledge.
• The mission frame 'Make AGI Accessible and Beneficial to Everyone' invokes a universal subject that erases situated, land-based knowledge systems.

• No acknowledgment of Indigenous or minority-nationality data communities (e.g. within China: Tibetan, Uyghur, Mongolian, Zhuang language and knowledge holders).
• No consent, benefit-sharing, or stewardship language around the text corpora used to train open-weight Yi models.
• Oral and non-textual traditions are absent from a text-centric, productivity-framed model of value.

Floor score. Nothing in the fetched copy touches Indigenous or community data sovereignty; the universalist productivity mission actively flattens the relational knowledge this lens asks after. Open weights do not substitute for sovereignty.

• One historically situated claim: 'excellent Chinese engineering heritage' names a national-cultural lineage rather than a placeless tech-universalism.
• CEO-level diplomacy (Kazakhstan President, Hong Kong Chief Executive's Council) gestures, implicitly, at the geopolitical economy 01.AI operates within.

• No transparency about the compute/GPU access constraints that materially shaped Yi's training (US export controls on advanced accelerators are a defining historical fact for a China-based lab and go unmentioned).
• No acknowledgment of colonial or extractive data legacies in the corpora.
• 'AI 2.0' framing presents a clean technological epoch, erasing the contingent labor, regulatory, and supply-chain history behind it.

Slightly above floor: the heritage and diplomacy framing show some historical self-location, but the most consequential historical force on this specific lab (export-control-shaped compute scarcity, which Kai-Fu Lee has discussed publicly) is invisible in the public copy. 'AI 2.0' is a temporally flat epoch-claim.

• Genuine bilingual (Chinese/English) substance is the lab's actual technical strength: Yi models are trained and benchmarked as bilingual rather than English-with-translation, which is more than token presence.
• The 'excellent Chinese engineering heritage' framing positions a non-Western locus of model development.

• Multilinguality in the public copy is asserted by focus area, not evidenced with named languages, scholar consultation, or preservation of culturally specific reasoning patterns.
• No engagement with the dozens of minority languages of China, let alone Global South languages.
• Reasoning is framed through a universal productivity logic; no recognition that categorical/argumentative structures differ across cultures.

Mid-low. Awarded above the marginalised-voices floor because 01.AI's bilingual Chinese/English capability is real and non-tokenistic (established public knowledge: Yi is a credibly bilingual open-weight family). But the homepage offers no on-page evidence of broader cultural plurality, and a productivity-universalist mission flattens difference.

• Open-weight Yi models are publicly available, which is the single most verification-friendly act a lab can take — third parties can inspect, replicate, and benchmark.
• A technical transparency channel is advertised ('Tech Blog'), and a commercial licensing framework is named, signalling some documentation discipline.

• The advertised Tech Blog (01-ai.github.io) returned HTTP 404 at audit time — the transparency pointer is broken, so claims could not be corroborated.
• No independent third-party audits of training data or bias are cited.
• No known-limitation disclosures, eval cards, or replication protocols appear in the fetched homepage copy.

Highest lens for this lab, and deservedly so on the strength of open weights (real, falsifiable verification surface). Capped at 5 because the homepage itself surfaces no limitation disclosures or external audits, and the one transparency link advertised was dead when fetched.

• The 'Human + AI' framing leaves a sliver of room for the human side of the dyad, which is marginally more affect-aware than pure automation copy.

• No acknowledgment of affective, intuitive, or aesthetic dimensions of intelligence.
• No space for ambiguity or poetic uncertainty; the register is engineering-and-enterprise throughout.
• Value is defined exclusively as 'productivity' and 'economic and societal values' — efficiency as the only mode of attention.

Near floor. The copy is instrumentally efficient and admits no feeling, ambiguity, or emotional labor. The 'Human +' fragment earns one point above the floor.

• The mission explicitly orients toward a future state ('AGI', 'AI 2.0'), so futures are nominally in view.
• 'Beneficial to Everyone' gestures at an inclusive future, even if undemonstrated.

• No engagement with labor displacement risk — striking given the copy's central claim is enhanced productivity, the exact mechanism that raises displacement questions.
• No environmental or compute-cost disclosure for training large foundation models.
• No democratic or participatory governance of agentic systems; the future is shaped at 'CEO-level' and via head-of-state engagement, i.e. elite-deliberated.

Low. Futures are asserted as benefit but the costs (labor, environment, power concentration) are unmodelled, and deliberation is explicitly top-down (CEO-level, head-of-state). 'Beneficial to Everyone' is a claim, not a participatory process.

• Open weights lower one access barrier for downstream developers, including under-resourced ones — a thin, indirect form of inclusion.

• No participatory design with Global South developers, no disability/accessibility commitments, no labor-representative engagement, no compensated feedback channels.
• The named stakeholders are exclusively powerful: presidents, chief executives, enterprise leadership.
• 'Everyone' is invoked rhetorically while the actual table seated in the copy contains only elites.

Floor score. The only voices present are heads of state and C-suites; the marginalised are named only via the empty universal 'Everyone.' Open weights earn the lab credit under scientific_evidence, not here, because access to weights is not participation.

• No irony, paradox, self-questioning, or willingness to name internal contradiction anywhere in the copy.

• The central contradiction — selling 'enhance human productivity' to enterprises while promising AGI 'Beneficial to Everyone' — is never acknowledged, let alone tested.
• The 'open-source' / 'commercial licensing framework' tension is presented as complementary rather than as a live conflict.
• The lab's own seriousness ('AI 2.0', 'AGI') is treated as exempt from scrutiny.

Floor score. Corporate mission copy is constitutionally trickster-free; here the solemnity of 'AGI / AI 2.0' is total and the open-vs-commercial and productivity-vs-everyone contradictions go entirely unnamed.