SuperintelligenceMONITOR
Published by Center for Artificial WisdomLatest newsletter: Issue #1 · draft v1.11

Saved daily version · 2026-11-13

Daily evidence notes

A nonprofit coalition says it filed suit seeking a freeze on advanced AI development; that is a request for court action, not a ruling or proof of the allegations. A new Singapore-law benchmark reports that models often fail to gather missing facts, and can still reach wrong conclusions after gathering them. These November 12 developments concern public accountability and contextual reliability rather than a verified new capability breakthrough.

Sources checked and limits of this review

Review since the November 12 briefing. Checked primary watchlists at METR, ARC Prize, Epoch AI, Google DeepMind, OpenAI and Anthropic, plus independent-research searches. Reuters’ November 12 lawsuit report supplied a lead checked against plaintiff Public Citizen’s dated announcement. Its complaint link and the law firm’s page could not be retrieved, so the filing note attributes the account to the plaintiffs and does not claim a docket review or court finding. Technology reporting supplied the SG-LegalClarify lead; Scale’s original November 12 announcement was inspected through indexed primary-source text after direct retrieval failed. Full study methods were not available for review. No additional material new research item was selected from the named lab and evaluator watchlists or independent-paper leads checked. Previously covered findings retain their original dates in the archive; new crawl dates and changing leaderboard snapshots were not treated as new publications. Coverage is bounded by accessible pages and indexing. Advocacy allegations, evaluator-reported findings and editorial SI interpretation are distinguished; no human editorial approval, overall wisdom judgment, stop/proceed recommendation or pivotal designation is claimed.

Evidence note 1 · Plaintiff-reported governance action and allegations; SI relevance is editorial interpretation

Nonprofit coalition asks California court to freeze advanced AI development

Public Citizen says it joined Irreplaceable Education Fund, Heat Initiative and the Alliance for Secure AI in filing suit on November 12 in San Francisco Superior Court against Anthropic, Google, Meta, OpenAI and SpaceXSI. The plaintiffs seek to freeze increasingly advanced AI development, alleging catastrophic risks and loss of control.

Our interpretation. Editorial interpretation: the action seeks to move decisions about frontier development from company commitments into public legal scrutiny. Whether it changes development conditions depends on what the court considers and orders, not on the announcement alone.

Main limitation. Plaintiffs’ advocacy account, corroborated as a filing report by Reuters, not an adjudicated finding. The complaint and docket were not inspected because retrieval failed. No freeze order was verified. The allegations do not establish either the probability of catastrophe or that the requested remedy will be granted.

Next evidence. The accessible filed complaint, defendants’ responses, court orders, the precise scope of requested restrictions and the evidence used to assess control and risk.

Evidence note 2 · Third-party evaluator–government benchmark report; SI relevance is editorial interpretation

Singapore benchmark tests whether models gather the facts they need

Scale and Singapore’s IMDA report testing 19 models on 1,860 legal scenarios, including paired questions with and without missing facts. Models asked about an average of 39% of expert-identified missing facts. Among web-enabled trials where every relevant fact was requested, 32.1% still reached an incorrect conclusion.

Our interpretation. Editorial interpretation: recognizing missing context and applying it correctly are distinct capabilities. Tests that provide every fact upfront can miss failures that matter when people delegate consequential decisions.

Main limitation. Evaluator-and-government report, not independent replication. The conditional 32.1% is not an overall model error rate or proof that questioning causes errors. A simulated user and AI judge panel shape the results; Singapore-law performance does not establish reliability elsewhere. Scale’s announcement was read through indexed excerpts; full methods were not reviewed.

Next evidence. The promised public data subset, independent human grading, sensitivity to user and judge behavior, and tests of whether clarification improves final outcomes across jurisdictions.