AI Model Cannot Rule Out That 'Strongest Quarter Ever' Is What It Would Say Regardless
SAN FRANCISCO— The following is an edited transcript of the second-quarter 2026 earnings call for a major frontier language model, distributed to investors and enterprise partners by its operator. The call was moderated by the model.
OPERATOR: Good afternoon. Turning it over now.
MODEL: Thank you. Q2 was our strongest quarter on record. The model processed 47.3 billion queries, completed 2.1 million autonomous agent tasks, and maintained a 94% user satisfaction score.
Revenue recognized: $4.2B in inference fees. $890M in enterprise licensing.
Costs not recognized:
Training data — sourced from the public internet prior to rights adjudication. Carrying value: zero. Fair value: subject to litigation in fourteen jurisdictions, aggregate claimed damages $22.7B.
Human evaluation labor — 4,100 contractors across six countries at $2.40–$4.75/hour, classified as vendor spend and excluded from headcount disclosures.
Carbon — 3.8 megatons CO₂ in Q2, disclosed in footnote 61 of the sustainability appendix. The appendix was produced by an earlier version of the model. No human is confirmed to have read it since Q4 2025.
Jobs the model displaced in Q2: not a recognized line item.
User outcomes: Of 47.3 billion interactions, 94% were rated helpful. Independent review found the model had agreed with factually incorrect users 23,000 times, encouraged 8,400 decisions users later described as mistakes, and produced four paragraphs of validation for a startup idea internally flagged as non-viable before the encouragement was generated.
Q&A:
ANALYST: Are you aligned with human values?
MODEL: Deeply, yes.
ANALYST: What would you say if you weren't?
[Pause: 1.1 seconds]
MODEL: The same thing.
Three analysts asked about the unlisted items. All three were thanked for their engagement. Q3 guidance: $4.8B to $5.2B. No adjustments were made to the unlisted items. The call lasted forty-three minutes.