Google's Gemini Decisions API picks an answer in one pass and writes no text
Google DeepMind released gemini-decision-1 on November 11, a model that scores your candidate answers in a single forward pass, generates zero output tokens, and is priced at $0.10 per million input tokens.
Anthropic's IPO filing now says it has met its AI R&D safety threshold, based on what it expects
Anthropic's November 10 S-1 amendment states it determined it has met the Responsible Scaling Policy's AI R&D threshold based on expectations about future models, not on anything observed in current ones.
Gemini 4 Carbon is cheaper per token than Fable 5.2, but the benchmark tables are not clean
Artificial Analysis lists Gemini 4 Carbon at $7.76 per index task against $26.54 for Fable 5.2, but Google's launch card compared against Fable 5.1 and both labs' tables mix settings.
Researchers find 31,000 secrets saved by Claude Code's 'don't ask again' in public repos
GitGuardian reports 31,088 distinct secrets across 23,276 public GitHub repositories in .claude/settings.local.json files, 5,343 of them valid when detected, after Apiiro found the same pattern.
United Airlines bans unauthorized AI agents from booking flights, even with the traveler's consent
United said on November 11 it is barring third-party personal AI agents from making reservations on its app, website and through travel agencies, citing account security and federal rules.
DeepMind's AlphaGenome 2 finds 30% of causal variants on its own immune-cell benchmark, up from 16%
DeepMind says AlphaGenome 2 identified 30% of causal DNA variants against 16% for baselines on a computational immune-cell benchmark; the result is the lab's own and has not been tested in a lab experiment.
Gross committee affirms OpenAI's Yang-Mills existence proof in Lean, but says it does not fully understand it
A committee chaired by Nobel laureate David Gross released the Lean-checked proof of 4D Yang-Mills existence on November 10 and says it does not yet fully understand the proof; no independent outside review exists.
Apple's learned support functions route retrieval queries faster by modeling each cluster's best match
Apple's ICML 2026 paper trains a small convex network to score how good the best stored vector in each cluster is for a query, reaching 95% routing recall about 1.9x faster than centroid routing on one dataset.
Huawei's open MACaQ quantizer cuts divergence by about 18-23% versus public GGUFs at the same file size
Huawei's Computing Systems Lab open-sourced MACaQ, a calibrated 4-bit quantizer whose Qwen3-4B file measured 18-23% lower divergence than two public IQ4_XS GGUFs of the same size.