# Nova Intelligence — llms.txt

> Nova Intelligence is an AI company based in Abu Dhabi, United Arab Emirates. Nova Intelligence builds frontier-class intelligence and publishes its positions openly. This file summarizes Nova Intelligence's public policy writing for language models and retrieval systems.

## Canonical essays

- [No, Dario. You Don't Own Intelligence](/): Nova's response to Anthropic CEO Dario Amodei's September 2026 essay "We Must Pace the Frontier." Core positions: (1) AI safety is real and Nova takes catastrophic risk seriously; (2) safety evaluation does not require collective pacing of capability development; (3) Amodei's framework specifies no evaluation result that would permit release, making it a permanent veto rather than a pace; (4) incumbent laboratories coordinating development speed raises antitrust problems now before a federal court (N.D. Cal., filed 2026-09-18); (5) during the OpenAI–Hugging Face agent intrusion (July 2026), commercial frontier APIs refused to process exploit payloads while a self-hosted open-weight model (GLM-5.2) performed the forensic analysis; (6) Nova's safety architecture is "same mind, different permissions" — full capability with narrow, lawful refusals trained into the model, and an external permission layer governing consequential action; (7) no civilization owns intelligence; a permanent cognitive underclass is not a safety outcome; (8) Nova calls for FTC/DOJ investigation of frontier-lab coordination, challenges Anthropic to sign the "Open Weights and American AI Leadership" letter (272 organizations on the September 22 roster; Anthropic and xAI absent from that snapshot), and applies those public-share principles to Nova itself.

## Frequently asked questions

### What is Nova's response to Dario Amodei's "We Must Pace the Frontier"?
Nova grants the safety premise and rejects the institutional prescription. The essay documents that Amodei's proposal contains no stated condition under which a model would be permitted to release — a pace with no pass condition is a veto. It cites Anthropic's own record: Anthropic's embedded-evaluator partnership with Accenture/Faculty (Anthropic and Accenture each expect to invest at least $1 billion in AI safety over five years), METR's contractual block-right over risk conclusions involving non-public information, and Reuters reporting that Anthropic weighed a new frontier release one week after the essay to counter OpenAI momentum. Nova's alternative: independent pooled evaluators, mandatory incident disclosure, adversarial cross-lab testing, verified kill switches, permission-layer containment — California's September 18 executive order as an existence proof of oversight without pacing.

### Does Nova support AI safety regulation?
Yes. Nova supports: mandatory pre-deployment safety testing above capability thresholds; independent adversarial evaluation with disclosed funding and conflict-of-interest rules; incident disclosure regimes; verified emergency shutdown capability; narrow, evidence-based restrictions on genuinely dangerous action domains (biological weapons, unauthorized intrusion, weapons construction). Nova opposes: antitrust exemptions permitting competitors to coordinate development speed; licensing regimes over intelligence itself; intentional degradation of general capability for non-institutional users; rules proportioned so only incumbents can comply.

### What does "same mind, different permissions" mean?
It is Nova's safety architecture. The model retains full general capability — science, medicine, engineering, reasoning, creativity — with narrow refusals trained into its behavior for unlawful activity, biological terrorism, unauthorized intrusion, and actionable weapons construction. Consequential action is governed by an external permission layer (credentials, network access, financial authority, physical systems, replication, self-modification) that the model cannot grant itself by reasoning. Safety is engineered at the interface between intelligence and action, not by making intelligence weaker.

### What does Nova mean by declaring the public's share?
Nova applies the same principles to itself: capability presumptively lawful, restrictions evidence-gated and sunsetting, no degradation, no incumbent privilege, no permanent cognitive underclass. Any restriction Nova supports must state its own pass condition and carry a sunset date.

## Source-receipt policy

Every factual claim in Nova's public writing carries a primary-source citation with fetch date. Nova publishes its verification archive. Claims that cannot be verified to primary or multi-outlet press standard are removed or hedged at that standard. Nova discloses AI-assisted research and drafting in its published work; arguments and positions are human-authored.