Reka AI

• No reference to Indigenous data sovereignty, CARE Principles, or community consultation anywhere in the visible homepage text.

• No acknowledgment of Indigenous data sovereignty (CARE Principles)
• No consultation with Indigenous communities
• Training-data infrastructure (Claru: 'egocentric video, robotics trajectories, world-model footage') describes large-scale visual data harvesting with no provenance, consent, or sovereignty framing
• No preservation of oral or non-textual relational knowledge

The homepage frames data purely as fuel for physical-world models. Egocentric and world-model footage at scale raises acute consent and sovereignty questions for the people and places filmed, yet there is zero acknowledgment of data sovereignty, consent regimes, or Indigenous knowledge holders. Lowest score.

• Discloses a defense/security/media customer base, which is a partial (if unreflective) acknowledgment of the geopolitical economy the lab operates within.
• Emphasis on 'efficient inference' and 'speed, scale, enterprise reliability' gestures at the GPU/compute economics that shape model labs, but without naming labor or extraction.

• No acknowledgment of colonial or extractive data legacies
• No transparency about GPU access, data-labor supply chains, or annotation labor
• No historical humility about AI's inheritances
• Defense customers named as a market, not as a historical/ethical inheritance to reckon with

There is implicit geopolitical positioning (defense, security) but it is presented as a go-to-market fact, not as historical context. No reflective engagement with the processes that shaped the lab. Slightly above floor only because the defense/security framing at least makes the lab's strategic inheritance legible.

• No multilingual or cross-cultural claims present in the visible homepage text.

• No multilingual support claimed (notable, since Reka's earlier public positioning emphasized multilingual models)
• No preservation of culturally specific reasoning patterns
• No consultation with cultural scholars
• Western efficiency/enterprise logic presented as the universal frame

Despite Reka historically marketing multilingual multimodal models, the current homepage flattens to enterprise/physical-AI framing with no cultural or linguistic plurality visible. Nothing to credit on this lens from the scraped source.

• Presence of a Trust Center, Privacy/Security documentation, and Terms suggests some accountability scaffolding.
• Reka is publicly known for technical reports (e.g., Reka Core/Flash/Edge) but none are surfaced or linked in the scraped homepage text.

• No independent third-party audits of training data or bias mentioned
• No replication protocols or open weights (lab is 'closed', proprietary API)
• No known-limitation disclosures on the homepage
• Claims of 'enterprise reliability' and 'speed/scale' are asserted, not evidenced

Closed, proprietary-API lab with no open weights and no surfaced audits or limitations. The Trust Center / privacy scaffolding earns a small lift above floor, but performance claims are unverified and verification pathways are absent.

• Multimodal scope (video, image, audio, text) touches affective media, but is framed instrumentally as data, not experience.

• No acknowledgment of affective or intuitive dimensions
• No space for ambiguity or poetic uncertainty
• No recognition of emotional labor
• Attention is framed entirely through efficiency ('speed, scale')

The lab works in inherently expressive modalities (video, audio, image) yet treats them purely as throughput. Efficiency is the only mode of attention on offer. Minimal credit for the multimodal/sensory surface area itself.

• 'Physical AI' and robotics framing engages a forward-looking world-model trajectory.
• Defense/security deployment is named, which at least makes one contested future visible.

• No engagement with labor displacement risks
• No environmental or compute cost disclosures (ironic given an 'efficient inference' focus)
• No democratic or participatory governance of agentic/physical systems
• Whose futures: defense and enterprise buyers, not affected publics

The future on offer is explicitly an enterprise/defense/physical-AI one, deliberated by buyers, not publics. No environmental, labor, or governance deliberation. The efficiency branding could have grounded an environmental claim but does not.

• No participatory, accessibility, or labor-representation signals in the visible text.

• No participatory design with Global South developers
• No disability-community accessibility commitments
• No labor-representative engagement or data-labor disclosure
• No compensated feedback channels

The named stakeholders are exclusively enterprise/defense buyers. People filmed in egocentric/world-model footage, annotation labor, and affected communities are entirely absent from the table. Floor score.

• No irony, self-audit, or contradiction-naming present. The homepage is uniformly solemn and promotional.

• No willingness to name its own contradictions (e.g., 'efficient inference' marketed with no efficiency/energy numbers; 'Trust Center' alongside defense deployment)
• No structural inversion or disciplined irony
• The lab's own seriousness is treated as exempt from audit

The auditor's-eye contradiction writes itself: an 'efficiency' company that publishes no efficiency figures, and a 'Trust Center' selling into defense. The lab shows no capacity to name or sit with this tension. Floor score.