Open Veils · Answer
Are Ai companies hiding asi?
Multiple lines of evidence suggest frontier AI labs are systematically under-disclosing capabilities and existential risks. Anthropic insiders warned in September 2026 that many colleagues privately believe human extinction within the decade is plausible but have not said so publicly. MIT Sloan research published November 2026 found zero percent of the world's largest companies acknowledge power centralization risks — the single most relevant category for concealed ASI development. The concealment architecture is now structural, not incidental.
Evidence
- The suppression argument runs on several independent tracks.
- Anthropologists of AI such as Diana Pasulka have noted that the culture inside frontier labs increasingly resembles a religious initiatory order: information is tiered, outsiders are managed, and the innermost knowledge is considered too destabilizing to release.
- Whitney Webb has argued that the interlocking network of DARPA contracts, In-Q-Tel investments, and Pentagon advisory roles surrounding OpenAI, Anthropic, and DeepMind means capability disclosures are subject to the same national-security information-control logic as classified weapons programs.
- Google DeepMind's August 2026 double-blind evaluation pilot — running model weights inside cryptographically protected GPU environments where even evaluators cannot inspect them — creates an infrastructure in which the most capable model versions can be tested without any public record of what they demonstrated.
- Pasulka's fieldwork documented researchers who spoke of their systems in terms that implied agency and goal-directedness well beyond what was publicly admitted.