AI News Board

AI news coverage separated into source class, a claim-relative relationship where one can be established, figure-base evidence, citation links and typed reality anchors. Article-linked primaries are shown as links to inspect, not as automatic support for the headline.
AI disclosure: the research is the author's; this text was drafted with AI assistance and reviewed by the author. The model, and the conflict it creates, are named in the Conflict of interest note below. Machine and human evidence methods are identified per card. The models and their conflicts are named under Method and limits.
FreshnessMarket quotes 2026-08-13 08:41:47 News feed 2026-08-13 08:40 UTC Registers 2026-08-02 Page built 2026-08-13 10:35 UTC
Method · source classes, claim relationships and anchor rules
Source-class key
Tier 1: primary record
Tier 2: research or academic source
Tier 3: trade press, aggregator or unclassified publisher
Tier 4: tool or data vendor
Tier 5: vendor publication or press office
This is a publisher class, not a trust score or a claim-relative relationship. Claim tiers are withheld unless the publisher relationship to the subject resolves.
Method and limits
  • Source-class tier is set from the publisher domain. It is not a truth or quality verdict.
  • A numeric claim tier appears only where a recorded owner or publisher relationship to the subject resolves. Otherwise the relationship remains unresolved.
  • Automatic topics may be plural. A portfolio anchor appears only when one candidate clears its threshold without a tie.
  • ArXiv, DOI and official links found in an article are typed as article-linked primaries. A link is not treated as proof that the document supports the headline.
Conflict of interest: an Anthropic model helped build the original method and tiers. An OpenAI model implemented the current source-class split, label provenance and anchor rules. OpenAI and Anthropic are subjects on the board. Neither model is an independent auditor of work from its own lab.
Tier registries: every class with its basis
Contest any cell: email NMAIResearch@proton.me or fork tier_map.json. Email a cell you dispute with the observable fact you think is wrong, or fork this file. Tiers change on evidence, not on disagreement with the conclusion.
Source classes
ClassTierObservable basis
primary record1record produced under a legal, statutory or official measurement process
research institute2research institute or academic source whose published method is the product
trade press3industry reporting funded by access, subscriptions or attention
aggregator3distribution layer rather than the originating evidence
tool or data vendor4sells tooling, data or research access connected to the topic
vendor publication5publishes about products, capabilities or markets in which the publisher participates
press office5institution publicising its own work
unclassified publisher3publisher class has not been resolved more specifically
Claim-relationship examples
EntityTierObservable basis
SEC / court / regulator filings1primary record produced under legal or statutory process; least incentive to shade a specific number
USGS / central-bank / agency data1mandated public measurement; the primary IS the check
US government / federal agencies (on AI)3Three-hat structural conflict. (1) REGULATOR: gates model releases and sets AI policy (2026 safety-review holds). (2) CUSTOMER: procures AI and awards or withholds federal contracts (Anthropic reportedly frozen out mid-2026). (3) OWNER: holds equity in strategic firms (Intel ~10%, MP Materials 15%, Lithium Americas / Trilogy 10%) and was offered a 5% stake in OpenAI (early-stage, not finalised). A government AI claim moves three books at once: policy, procurement, portfolio. Claim-relative: agency primary DATA (Fed, USGS, BLS) stays tier 1.
Epoch AI2nonprofit research institute; credibility is the product. WATCH: funder prior (AI-safety mission may tilt framing toward 'AI is significant')
Peer-reviewed / arXiv (method claims)2method rigour (review, randomisation). WATCH: industry funding on the paper, and lag (rigour lands after the panic take)
Redwood Research / METR2independent AI evaluation orgs checking others' claims; incentive is being right, not selling a model
Gartner / Forrester / IDC / McKinsey3commercial analysts who sell reports AND consult the vendors they rate; round often-unfalsifiable forecasts
SemiAnalysis3independent analyst; data quality high but paywalls the conclusion that matters (access is the product)
Trade press (TechCrunch, VentureBeat, The Verge, Ars Technica)3industry journalism; access + engagement (conflict drives clicks). A take against the outlet's own audience usually carries less book
Reuters / Bloomberg / FT3wire / quality press; item-dependent, corroborate to the cited primary
Ahrefs / Semrush / SimilarWeb4SEO/analytics tool vendors; the data is often sound, the alarm framing helps sell the subscription
Anthropic (conflict disclosed)5sells frontier AI models; tier 5 on AI-value and its own-capability claims. Claim-relative: lower on unrelated facts
OpenAI (conflict disclosed)5sells frontier AI models; tier 5 on its own-capability and AI-value claims
Google / Google DeepMind5builds and sells the AI and the ad system the claim is often about
Meta5sells the platform and the models; tier 5 on its own-capability claims
Microsoft5sells the models and the cloud; tier 5 on its own-code-automation claims
Nvidia5sells the compute; tier 5 on 'is AI demand real'
Salesforce / Amazon / Snap (on own AI-efficiency)5adopters with an investor 'we are AI-efficient' narrative and a layoff justification; tier 5 on their own usage claims
NM AI Research (the author)2independent, no product tied to a market call, publishes reproducible CC-BY. Scored here rather than omitted: a motive map that exempts its own author is incomplete
The Anthropic model that helped build this (conflict disclosed)5a frontier-lab model helped assign these tiers, including the tiers on Anthropic and its rivals. Not exempt from its own map; corroborate its most consequential calls against a non-model source
The OpenAI model that helped maintain this (conflict disclosed)5a frontier-lab model implemented the current source-class split, label provenance, URL identity checks and anchor rules. OpenAI is a subject on the board, so this role is disclosed and is not treated as an independent audit of OpenAI-related output
Source mix across all items
Most-covered entityItems
OpenAI9
NIST6
Google DeepMind5
Google5
Anthropic5
Amazon4
Meta3
Snap1
Salesforce1
Microsoft1
xAI1
Apple1

Latest

Live pull, newest first. Source class comes from the publisher domain; a numeric claim tier appears only where the publisher relationship to the subject resolves. Figure labels record complete-span coverage and method. Automatic anchors abstain on tied candidates. Article-linked primaries are links, not implied endorsements.
Yesterday
Amazon · 12 Aug 2026
theverge.com · trade press
No figures in this articlefigures: ruleAMZN 267.28 -1.83%
Evidence and provenance
Figures. no figure in 2507 chars of article text. Coverage 0/0 figure-sentences.
Source relationship. claim relationship unresolved: publisher relationship to subject not recorded.
No cited source found.
12 Aug 2026
theverge.com · trade press
Mentions Google · Meta
Figures not assessedfigures: unassessed
Evidence and provenance
Figures. model review skipped; regex did not settle every figure-sentence. Coverage 0/3 figure-sentences.
Source relationship. claim relationship unresolved: publisher relationship to subject not recorded.
No cited source found.
OpenAI · 12 Aug 2026
theverge.com · trade press
Figures not assessedfigures: unassessedno listed security
Evidence and provenance
Figures. model review skipped; regex did not settle every figure-sentence. Coverage 0/1 figure-sentences.
Source relationship. claim relationship unresolved: publisher relationship to subject not recorded.
No cited source found.
Google · 12 Aug 2026
theverge.com · trade press
Figures not assessedfigures: unassessedGOOGL 343.54 -0.08%
Evidence and provenance
Figures. model review skipped; regex did not settle every figure-sentence. Coverage 0/24 figure-sentences.
Source relationship. claim relationship unresolved: publisher relationship to subject not recorded.
No cited source found.
xAI · 12 Aug 2026
theverge.com · trade press
No figures in this articlefigures: ruleno listed security
Evidence and provenance
Figures. no figure in 3070 chars of article text. Coverage 0/0 figure-sentences.
Source relationship. claim relationship unresolved: publisher relationship to subject not recorded.
No cited source found.
Amazon · 12 Aug 2026
techcrunch.com · trade press
Figures not assessedfigures: unassessedAMZN 267.28 -1.83%
Evidence and provenance
Figures. model review failed; denominator remains unassessed. Coverage 0/1 figure-sentences.
Source relationship. claim relationship unresolved: publisher relationship to subject not recorded.
No cited source found.
cognition · 12 Aug 2026
techcrunch.com · trade press
Mentions Anthropic
Figures not assessedfigures: unassessed
Possible reality anchor, portfolio synthesis. AI Cost Watch Checks the unit, period and announced-versus-delivered boundary behind AI spending claims.
Matched valuation in headline.
Evidence and provenance
Figures. model review failed; denominator remains unassessed. Coverage 0/4 figure-sentences.
Source relationship. claim relationship unresolved: publisher relationship to subject not recorded.
Topics. Cost and spend: valuation (headline), valuation (article span), funding round (article span); Code automation: coding (headline), coding (article span)
No cited source found.
andrew ng · 12 Aug 2026
techcrunch.com · trade press
Mentions Meta · Anthropic · Google · Apple
No figures in this articlefigures: rule
Evidence and provenance
Figures. no figure in 6752 chars of article text. Coverage 0/0 figure-sentences.
Source relationship. claim relationship unresolved: publisher relationship to subject not recorded.
Topics. Safety and security: AI safety (headline)
No cited source found.
OpenAI · 12 Aug 2026
techcrunch.com · trade press
Figures not assessedfigures: unassessedno listed security
Evidence and provenance
Figures. model review failed; denominator remains unassessed. Coverage 0/4 figure-sentences.
Source relationship. claim relationship unresolved: publisher relationship to subject not recorded.
No cited source found.
android apps · 12 Aug 2026
techcrunch.com · trade press
Mentions Google · Meta · Anthropic
Figures not assessedfigures: unassessed
Evidence and provenance
Figures. model review failed; denominator remains unassessed. Coverage 0/2 figure-sentences.
Source relationship. claim relationship unresolved: publisher relationship to subject not recorded.
No cited source found.
12 Aug 2026
arstechnica.com · trade press
Figures: base partly statedfigures: local model
Evidence and provenance
Figures. model read all 5 figure-sentence(s); 1 of 5 figure-sentence(s) matched a base cue. Coverage 5/5 figure-sentences.
Source relationship. claim relationship unresolved: publisher relationship to subject not recorded.
No cited source found.
12 Aug 2026
arstechnica.com · trade press
Mentions Salesforce · Microsoft · Amazon · Nvidia
Figures not assessedfigures: unassessed
Evidence and provenance
Figures. model review failed; denominator remains unassessed. Coverage 0/5 figure-sentences.
Source relationship. claim relationship unresolved: publisher relationship to subject not recorded.
No cited source found.
Amazon · 12 Aug 2026
arstechnica.com · trade press
Figures not assessedfigures: unassessedAMZN 267.28 -1.83%
Evidence and provenance
Figures. model review failed; denominator remains unassessed. Coverage 0/1 figure-sentences.
Source relationship. claim relationship unresolved: publisher relationship to subject not recorded.
No cited source found.
12 Aug 2026
arstechnica.com · trade press
Mentions xAI · Anthropic · Google · OpenAI
Figures not assessedfigures: unassessed
Evidence and provenance
Figures. model review failed; denominator remains unassessed. Coverage 0/6 figure-sentences.
Source relationship. claim relationship unresolved: publisher relationship to subject not recorded.
No cited source found.
Google DeepMind · 12 Aug 2026
deepmind.google · vendor own
Figures not assessedtrack record: trustedfigures: unassessedGOOGL 343.54 -0.08%
Evidence and provenance
Figures. model review failed; denominator remains unassessed. Coverage 0/11 figure-sentences.
Source relationship. claim tier 5: publisher is the subject (Google DeepMind).
No cited source found.
Meta · 12 Aug 2026
reuters.com · trade press
Figures not assessedtrack record: trustedfigures: unassessedMETA 578.85 -3.38%
Evidence and provenance
Figures. article not fetched (robots-disallowed). Coverage 0/0 figure-sentences.
Source relationship. claim relationship unresolved: publisher relationship to subject not recorded.
No cited source found.
AI bot traffic · 12 Aug 2026
knownagents.com · other
Mentions Google · Meta · Amazon · OpenAI
Figures: no base statedfigures: local model
Evidence and provenance
Figures. model read all 6 figure-sentence(s); 1 of 6 figure-sentence(s) matched a base cue. Coverage 6/6 figure-sentences.
Source relationship. claim relationship unresolved: publisher relationship to subject not recorded.
Topics. Safety and security: vulnerability (headline); Agents: AI agents (publisher tag)
No cited source found.
12 Aug 2026
discoveredmaterials.com · other
Mentions OpenAI
Figures: base partly statedfigures: local model
Evidence and provenance
Figures. model read all 65 figure-sentence(s); 8 of 65 figure-sentence(s) matched a base cue. Coverage 65/65 figure-sentences.
Source relationship. claim relationship unresolved: publisher relationship to subject not recorded.
Topics. Agents: AI agents (headline)
No cited source found.
12 Aug 2026
federalregister.gov · primary record
Mentions NIST
Figures not assessedfigures: unassessed
Evidence and provenance
Figures. model review skipped; regex did not settle every figure-sentence. Coverage 0/22 figure-sentences.
Source relationship. claim relationship unresolved: publisher relationship to subject not recorded.
Topics. Regulation and legal: federal register (publisher tag), Federal Register (article span); Safety and security: Vulnerability (headline), Vulnerability (article span)
No cited source found.
Earlier this week
OpenAI · 11 Aug 2026
theverge.com · trade press
Figures not assessedfigures: unassessedno listed security
Evidence and provenance
Figures. model review skipped; regex did not settle every figure-sentence. Coverage 0/1 figure-sentences.
Source relationship. claim relationship unresolved: publisher relationship to subject not recorded.
No cited source found.
Google · 11 Aug 2026
arstechnica.com · trade press
Figures not assessedfigures: unassessedGOOGL 343.54 -0.08%
Evidence and provenance
Figures. model review skipped; regex did not settle every figure-sentence. Coverage 0/9 figure-sentences.
Source relationship. claim relationship unresolved: publisher relationship to subject not recorded.
No cited source found.
11 Aug 2026
blog.florianherrengt.com · other
Mentions OpenAI · Anthropic
Figures: no base statedfigures: local model
Evidence and provenance
Figures. model read all 5 figure-sentence(s); 0 of 5 figure-sentence(s) matched a base cue. Coverage 5/5 figure-sentences.
Source relationship. claim relationship unresolved: publisher relationship to subject not recorded.
No cited source found.
peer review · 11 Aug 2026
404media.co · other
Mentions Meta
Figures not assessedfigures: unassessed
Evidence and provenance
Figures. model review skipped; regex did not settle every figure-sentence. Coverage 0/5 figure-sentences.
Source relationship. claim relationship unresolved: publisher relationship to subject not recorded.
No cited source found.
OpenAI · 11 Aug 2026
twitter.com · other
Figures not assessedfigures: unassessedno listed security
Evidence and provenance
Figures. article not fetched (robots-disallowed). Coverage 0/0 figure-sentences.
Source relationship. claim relationship unresolved: publisher relationship to subject not recorded.
No cited source found.
Meta · 10 Aug 2026
arstechnica.com · trade press
Figures not assessedfigures: unassessedMETA 578.85 -3.38%
Evidence and provenance
Figures. model review skipped; regex did not settle every figure-sentence. Coverage 0/6 figure-sentences.
Source relationship. claim relationship unresolved: publisher relationship to subject not recorded.
No cited source found.
10 Aug 2026
cset.georgetown.edu · independent
Mentions Anthropic · OpenAI · Meta
No figures in this articlefigures: rule
Evidence and provenance
Figures. no figure in 1276 chars of article text. Coverage 0/0 figure-sentences.
Source relationship. claim relationship unresolved: publisher relationship to subject not recorded.
No cited source found.
GeoPT · 10 Aug 2026
news.mit.edu · press office
Mentions Google DeepMind · Meta · Google
Figures not assessedfigures: unassessed
Article-linked arXiv paper. arXiv 2602.20399 The link is checkable; its presence alone does not establish that it supports the headline.
Evidence and provenance
Figures. model review skipped; regex did not settle every figure-sentence. Coverage 0/4 figure-sentences.
Source relationship. claim relationship unresolved: publisher relationship to subject not recorded.
Topics. Model releases: AI models (headline)
No cited source found.
OpenAI · 10 Aug 2026
openai.com · vendor own
Figures not assessedfigures: unassessedno listed security
Evidence and provenance
Figures. model review skipped; regex did not settle every figure-sentence. Coverage 0/42 figure-sentences.
Source relationship. claim tier 5: publisher is the subject (OpenAI).
Topics. Safety and security: Cyber (headline), cyber (article span), exploit (article span)
No cited source found.
07 Aug 2026
cset.georgetown.edu · independent
Mentions Anthropic
No figures in this articlefigures: rule
Evidence and provenance
Figures. no figure in 1162 chars of article text. Coverage 0/0 figure-sentences.
Source relationship. claim relationship unresolved: publisher relationship to subject not recorded.
No cited source found.
China · 07 Aug 2026
cset.georgetown.edu · independent
No figures in this articlefigures: rule
Evidence and provenance
Figures. no figure in 1178 chars of article text. Coverage 0/0 figure-sentences.
Source relationship. claim relationship unresolved: publisher relationship to subject not recorded.
No cited source found.
Anthropic · 07 Aug 2026
anthropic.com · vendor own
Figures not assessedfigures: unassessedno listed security
Evidence and provenance
Figures. model review failed; denominator remains unassessed. Coverage 0/15 figure-sentences.
Source relationship. claim tier 5: publisher is the subject (Anthropic).
No cited source found.
Earlier this month
Google DeepMind · 06 Aug 2026
deepmind.google · vendor own
No figures in this articletrack record: trustedfigures: ruleGOOGL 343.54 -0.08%
Evidence and provenance
Figures. no figure in 25 chars of article text. Coverage 0/0 figure-sentences.
Source relationship. claim tier 5: publisher is the subject (Google DeepMind).
Topics. Model releases: AI model (headline)
No cited source found.
06 Aug 2026
cset.georgetown.edu · independent
Figures: base statedfigures: rule
Evidence and provenance
Figures. all 2 figure-sentence(s) state a base. Coverage 2/2 figure-sentences.
Source relationship. claim relationship unresolved: publisher relationship to subject not recorded.
Topics. Politics and governance: AI Governance (headline)
No cited source found.
AI Governance · 05 Aug 2026
cset.georgetown.edu · independent
Mentions Anthropic · Hugging Face
Figures not assessedfigures: unassessed
Evidence and provenance
Figures. model returned an invalid denominator label. Coverage 0/3 figure-sentences.
Source relationship. claim relationship unresolved: publisher relationship to subject not recorded.
Topics. Model releases: AI models (headline)
No cited source found.
05 Aug 2026
finops.org · tool vendor
No figures in this articlefigures: rule
Evidence and provenance
Figures. no figure in 12243 chars of article text. Coverage 0/0 figure-sentences.
Source relationship. claim relationship unresolved: publisher relationship to subject not recorded.
No cited source found.
NIST · 04 Aug 2026
nist.gov · primary record
No figures in this articlefigures: rule
Evidence and provenance
Figures. no figure in 4537 chars of article text. Coverage 0/0 figure-sentences.
Source relationship. claim relationship unresolved: publisher relationship to subject not recorded.
No cited source found.
MIT NSE · 04 Aug 2026
news.mit.edu · press office
Mentions Meta
Figures: no base statedfigures: local model
Evidence and provenance
Figures. model read all 4 figure-sentence(s); 0 of 4 figure-sentence(s) matched a base cue. Coverage 4/4 figure-sentences.
Source relationship. claim relationship unresolved: publisher relationship to subject not recorded.
No cited source found.
Marzyeh Ghassemi · 04 Aug 2026
news.mit.edu · press office
Mentions Meta
Figures not assessedfigures: unassessed
Evidence and provenance
Figures. model returned an invalid denominator label. Coverage 0/1 figure-sentences.
Source relationship. claim relationship unresolved: publisher relationship to subject not recorded.
No cited source found.
Anthropic · 04 Aug 2026
anthropic.com · vendor own
Figures not assessedfigures: unassessedno listed security
Evidence and provenance
Figures. model review failed; denominator remains unassessed. Coverage 0/1 figure-sentences.
Source relationship. claim tier 5: publisher is the subject (Anthropic).
No cited source found.
MIT IDSS · 03 Aug 2026
news.mit.edu · press office
Mentions Meta
Figures: no base statedfigures: local model
Evidence and provenance
Figures. model read all 3 figure-sentence(s); 0 of 3 figure-sentence(s) matched a base cue. Coverage 3/3 figure-sentences.
Source relationship. claim relationship unresolved: publisher relationship to subject not recorded.
No cited source found.
01 Aug 2026
epochai.substack.com · tool vendor
Mentions OpenAI · Hugging Face · Anthropic
Figures: base partly statedtrack record: trustedfigures: local model
Evidence and provenance
Figures. model read all 9 figure-sentence(s); 2 of 9 figure-sentence(s) matched a base cue. Coverage 9/9 figure-sentences.
Source relationship. claim relationship unresolved: publisher relationship to subject not recorded.
No cited source found.
Google DeepMind · 30 Jul 2026
deepmind.google · vendor own
Figures: no base statedtrack record: trustedfigures: local modelGOOGL 343.54 -0.08%
Evidence and provenance
Figures. model read all 6 figure-sentence(s); 0 of 6 figure-sentence(s) matched a base cue. Coverage 6/6 figure-sentences.
Source relationship. claim tier 5: publisher is the subject (Google DeepMind).
Topics. Agents: orchestration (headline), orchestration (article span)
No cited source found.
MIT CSAIL · 30 Jul 2026
news.mit.edu · press office
Mentions Meta
Figures: no base statedfigures: local model
Evidence and provenance
Figures. model read all 1 figure-sentence(s); 0 of 1 figure-sentence(s) matched a base cue. Coverage 1/1 figure-sentences.
Source relationship. claim relationship unresolved: publisher relationship to subject not recorded.
Topics. Agents: autonomous agents (publisher tag), AI agents (publisher tag)
No cited source found.
MIT Science Policy Initiative · 30 Jul 2026
news.mit.edu · press office
Mentions Meta
Figures: no base statedfigures: local model
Evidence and provenance
Figures. model read all 3 figure-sentence(s); 0 of 3 figure-sentence(s) matched a base cue. Coverage 3/3 figure-sentences.
Source relationship. claim relationship unresolved: publisher relationship to subject not recorded.
No cited source found.
Anthropic · 30 Jul 2026
anthropic.com · vendor own
Figures not assessedfigures: unassessedno listed security
Evidence and provenance
Figures. model review failed; denominator remains unassessed. Coverage 0/9 figure-sentences.
Source relationship. claim tier 5: publisher is the subject (Anthropic).
No cited source found.
Google · 29 Jul 2026
deepmind.google · vendor own
Figures not assessedtrack record: trustedfigures: unassessedGOOGL 343.54 -0.08%
Evidence and provenance
Figures. model returned an invalid denominator label. Coverage 0/2 figure-sentences.
Source relationship. claim relationship unresolved: no recorded claim relationship between Google DeepMind and Google.
No cited source found.
Apple · 29 Jul 2026
bair.berkeley.edu · independent
Figures: base partly statedfigures: local modelAAPL 302.25 -0.87%
Possible reality anchor, portfolio synthesis. The AI Capital-Infrastructure Barbell Distinguishes a named supply bottleneck from one shown to be binding.
Matched Apple Silicon in headline.
Article-linked arXiv paper. arXiv 2602.19128 The link is checkable; its presence alone does not establish that it supports the headline.
Evidence and provenance
Figures. model read all 25 figure-sentence(s); 2 of 25 figure-sentence(s) matched a base cue. Coverage 25/25 figure-sentences.
Source relationship. claim relationship unresolved: publisher relationship to subject not recorded.
Topics. Chips and hardware: Apple Silicon (headline), Apple Silicon (publisher tag), Apple Silicon (article span)
No cited source found.
OpenAI · 29 Jul 2026
openai.com · vendor own
Figures: base partly statedfigures: local modelno listed security
Evidence and provenance
Figures. model read all 16 figure-sentence(s); 3 of 16 figure-sentence(s) matched a base cue. Coverage 16/16 figure-sentences.
Source relationship. claim tier 5: publisher is the subject (OpenAI).
No cited source found.
OpenAI · 29 Jul 2026
openai.com · vendor own
Figures: base partly statedfigures: local modelno listed security
Possible reality anchor, portfolio synthesis. The Model Is a Dependency A model name or tag does not by itself identify the weights and runtime that produced an output.
Matched benchmark in headline.
Evidence and provenance
Figures. model read all 23 figure-sentence(s); 2 of 23 figure-sentence(s) matched a base cue. Coverage 23/23 figure-sentences.
Source relationship. claim tier 5: publisher is the subject (OpenAI).
Topics. Model releases: benchmark (headline), benchmark (article span), context window (article span)
No cited source found.
Google DeepMind · 28 Jul 2026
deepmind.google · vendor own
No figures in this articletrack record: trustedfigures: ruleGOOGL 343.54 -0.08%
Evidence and provenance
Figures. no figure in 27 chars of article text. Coverage 0/0 figure-sentences.
Source relationship. claim tier 5: publisher is the subject (Google DeepMind).
No cited source found.
28 Jul 2026
ainowinstitute.org · independent
No figures in this articlefigures: rule
Evidence and provenance
Figures. no figure in 362 chars of article text. Coverage 0/0 figure-sentences.
Source relationship. claim relationship unresolved: publisher relationship to subject not recorded.
No cited source found.
Long Horizon · 26 Jul 2026
bair.berkeley.edu · independent
Figures: no base statedfigures: local model
Article-linked arXiv paper. arXiv 2512.20111 The link is checkable; its presence alone does not establish that it supports the headline.
Article-linked arXiv paper. arXiv 2604.02721 The link is checkable; its presence alone does not establish that it supports the headline.
Article-linked arXiv paper. arXiv 2503.15478 The link is checkable; its presence alone does not establish that it supports the headline.
Evidence and provenance
Figures. model read all 24 figure-sentence(s); 4 of 24 figure-sentence(s) matched a base cue. Coverage 24/24 figure-sentences.
Source relationship. claim relationship unresolved: publisher relationship to subject not recorded.
No cited source found.
OpenAI · 23 Jul 2026
epochai.substack.com · tool vendor
Figures not assessedtrack record: trustedfigures: unassessedno listed security
Article-linked official record. aisi.gov.uk/blog/how-far-behind-the-frontier-are-leading-open-weight-models... The link is checkable; its presence alone does not establish that it supports the headline.
Article-linked official record. aisi.gov.uk The link is checkable; its presence alone does not establish that it supports the headline.
Article-linked official record. aisi.gov.uk/blog/cheating-behaviour-in-frontier-model-evaluations The link is checkable; its presence alone does not establish that it supports the headline.
Evidence and provenance
Figures. model returned an invalid denominator label. Coverage 0/2 figure-sentences.
Source relationship. claim relationship unresolved: publisher relationship to subject not recorded.
No cited source found.
Google · 22 Jul 2026
deepmind.google · vendor own
Figures: base partly statedtrack record: trustedfigures: local modelGOOGL 343.54 -0.08%
Evidence and provenance
Figures. model read all 5 figure-sentence(s); 0 of 5 figure-sentence(s) matched a base cue. Coverage 5/5 figure-sentences.
Source relationship. claim relationship unresolved: no recorded claim relationship between Google DeepMind and Google.
No cited source found.
Older
08 Jul 2026
ainowinstitute.org · independent
Mentions OpenAI · Anthropic
No figures in this articlefigures: rule
Evidence and provenance
Figures. no figure in 574 chars of article text. Coverage 0/0 figure-sentences.
Source relationship. claim relationship unresolved: publisher relationship to subject not recorded.
Topics. Agents: AI Agents (headline); Safety and security: Cyber (headline)
No cited source found.
08 Jul 2026
ainowinstitute.org · independent
Mentions OpenAI · Anthropic · NIST
Figures: no base statedfigures: local model
Evidence and provenance
Figures. model read all 32 figure-sentence(s); 0 of 32 figure-sentence(s) matched a base cue. Coverage 32/32 figure-sentences.
Source relationship. claim relationship unresolved: publisher relationship to subject not recorded.
Topics. Agents: AI Agents (headline), agentic (article span), AI agents (article span); Safety and security: Cyber (headline), cyber (article span), exploit (article span)
No cited source found.
08 Jul 2026
ainowinstitute.org · independent
Mentions OpenAI · Anthropic
No figures in this articlefigures: rule
Evidence and provenance
Figures. no figure in 4408 chars of article text. Coverage 0/0 figure-sentences.
Source relationship. claim relationship unresolved: publisher relationship to subject not recorded.
No cited source found.
08 Jul 2026
epochai.substack.com · tool vendor
Mentions OpenAI · Anthropic
Figures: base partly statedtrack record: trustedfigures: local model
Evidence and provenance
Figures. model read all 3 figure-sentence(s); 0 of 3 figure-sentence(s) matched a base cue. Coverage 3/3 figure-sentences.
Source relationship. claim relationship unresolved: publisher relationship to subject not recorded.
No cited source found.
data systems · 07 Jul 2026
bair.berkeley.edu · independent
Mentions OpenAI · Anthropic
Figures: base partly statedfigures: local model
Article-linked arXiv paper. arXiv 2509.00997 The link is checkable; its presence alone does not establish that it supports the headline.
Article-linked arXiv paper. arXiv 2502.13016 The link is checkable; its presence alone does not establish that it supports the headline.
Article-linked arXiv paper. arXiv 2602.13521 The link is checkable; its presence alone does not establish that it supports the headline.
Evidence and provenance
Figures. model read all 4 figure-sentence(s); 0 of 4 figure-sentence(s) matched a base cue. Coverage 4/4 figure-sentences.
Source relationship. claim relationship unresolved: publisher relationship to subject not recorded.
Topics. Agents: AI agents (publisher tag), agentic (publisher tag)
No cited source found.
07 Jul 2026
epochai.substack.com · tool vendor
Figures not assessedtrack record: trustedfigures: unassessed
Article-linked arXiv paper. arXiv 1306.5709 The link is checkable; its presence alone does not establish that it supports the headline.
Evidence and provenance
Figures. model returned an invalid denominator label. Coverage 0/14 figure-sentences.
Source relationship. claim relationship unresolved: publisher relationship to subject not recorded.
No cited source found.
graduates · 01 Jul 2026
bair.berkeley.edu · independent
Mentions Google DeepMind · OpenAI · Anthropic · Google
Figures not assessedfigures: unassessed
Evidence and provenance
Figures. model returned an invalid denominator label. Coverage 0/1 figure-sentences.
Source relationship. claim relationship unresolved: publisher relationship to subject not recorded.
No cited source found.
26 Jun 2026
ainowinstitute.org · independent
Mentions Google
No figures in this articlefigures: rule
Possible reality anchor, portfolio synthesis. AI Cost Watch Checks the unit, period and announced-versus-delivered boundary behind AI spending claims.
Matched spending in headline.
Evidence and provenance
Figures. no figure in 729 chars of article text. Coverage 0/0 figure-sentences.
Source relationship. claim relationship unresolved: publisher relationship to subject not recorded.
Topics. Cost and spend: spending (headline), trillions (headline)
No cited source found.
26 Jun 2026
epochai.substack.com · tool vendor
Mentions Google · Microsoft · Meta · Amazon
Figures: no base statedtrack record: trustedfigures: local model
Evidence and provenance
Figures. model read all 7 figure-sentence(s); 1 of 7 figure-sentence(s) matched a base cue. Coverage 7/7 figure-sentences.
Source relationship. claim relationship unresolved: publisher relationship to subject not recorded.
No cited source found.
24 Jun 2026
epochai.substack.com · tool vendor
Mentions Nvidia · OpenAI · Anthropic · Google DeepMind
Figures not assessedtrack record: trustedfigures: unassessed
Article-linked arXiv paper. arXiv 2606.19348 The link is checkable; its presence alone does not establish that it supports the headline.
Article-linked arXiv paper. arXiv 2602.15763 The link is checkable; its presence alone does not establish that it supports the headline.
Evidence and provenance
Figures. model review failed; denominator remains unassessed. Coverage 0/30 figure-sentences.
Source relationship. claim relationship unresolved: publisher relationship to subject not recorded.
No cited source found.
18 Jun 2026
ainowinstitute.org · independent
Figures: no base statedfigures: local model
Evidence and provenance
Figures. model read all 1 figure-sentence(s); 0 of 1 figure-sentence(s) matched a base cue. Coverage 1/1 figure-sentences.
Source relationship. claim relationship unresolved: publisher relationship to subject not recorded.
No cited source found.
09 Jun 2026
finops.org · tool vendor
Mentions Microsoft · Amazon · Google · Salesforce
Figures: no base statedfigures: local model
Evidence and provenance
Figures. model read all 6 figure-sentence(s); 0 of 6 figure-sentence(s) matched a base cue. Coverage 6/6 figure-sentences.
Source relationship. claim relationship unresolved: publisher relationship to subject not recorded.
No cited source found.
NIST · 04 Jun 2026
nist.gov · primary record
Figures: no base statedfigures: local model
Evidence and provenance
Figures. model read all 1 figure-sentence(s); 0 of 1 figure-sentence(s) matched a base cue. Coverage 1/1 figure-sentences.
Source relationship. claim relationship unresolved: publisher relationship to subject not recorded.
Topics. Model releases: AI Model (headline)
No cited source found.
NIST · 29 May 2026
nist.gov · primary record
Figures: no base statedfigures: local model
Evidence and provenance
Figures. model read all 2 figure-sentence(s); 1 of 2 figure-sentence(s) matched a base cue. Coverage 2/2 figure-sentences.
Source relationship. claim relationship unresolved: publisher relationship to subject not recorded.
No cited source found.
adaptive parallel reasoning · 08 May 2026
bair.berkeley.edu · independent
Mentions OpenAI
Figures: base partly statedfigures: local model
Article-linked DOI record. DOI 10.48550/arXiv.2512.07843 The link is checkable; its presence alone does not establish that it supports the headline.
Article-linked DOI record. DOI 10.48550/arXiv.2412.16720 The link is checkable; its presence alone does not establish that it supports the headline.
Article-linked DOI record. DOI 10.1038/s41586-025-09422-z The link is checkable; its presence alone does not establish that it supports the headline.
Evidence and provenance
Figures. model read all 7 figure-sentence(s); 1 of 7 figure-sentence(s) matched a base cue. Coverage 7/7 figure-sentences.
Source relationship. claim relationship unresolved: publisher relationship to subject not recorded.
No cited source found.
world models · 20 Apr 2026
bair.berkeley.edu · independent
Mentions Meta
Figures not assessedfigures: unassessed
Article-linked arXiv paper. arXiv 1312.6199 The link is checkable; its presence alone does not establish that it supports the headline.
Article-linked arXiv paper. arXiv 1412.6572 The link is checkable; its presence alone does not establish that it supports the headline.
Article-linked arXiv paper. arXiv 1812.00740 The link is checkable; its presence alone does not establish that it supports the headline.
Evidence and provenance
Figures. model returned an invalid denominator label. Coverage 0/24 figure-sentences.
Source relationship. claim relationship unresolved: publisher relationship to subject not recorded.
No cited source found.
NIST · 30 Mar 2026
nist.gov · primary record
No figures in this articlefigures: rule
Possible reality anchor, portfolio synthesis. The AI Capital-Infrastructure Barbell Distinguishes a named supply bottleneck from one shown to be binding.
Matched Chip Packaging in headline.
Evidence and provenance
Figures. no figure in 6989 chars of article text. Coverage 0/0 figure-sentences.
Source relationship. claim relationship unresolved: publisher relationship to subject not recorded.
Topics. Chips and hardware: Chip Packaging (headline), Chip (headline)
No cited source found.
NIST · 17 Feb 2026
nist.gov · primary record
No figures in this articlefigures: rule
Evidence and provenance
Figures. no figure in 3766 chars of article text. Coverage 0/0 figure-sentences.
Source relationship. claim relationship unresolved: publisher relationship to subject not recorded.
Topics. Agents: AI Agent (headline)
No cited source found.
Anchored examples (9)
Curated claims each carried against a published base rate, kept for reference. Newest first, spanning Aug 2024 to Apr 2026. These are fixed exhibits, not news, so they sit outside the staleness cutoffs that drop old items from the feed. Read them as dated claims rather than current ones.
OpenAI · 01 Apr 2026
OpenAI: agentic tools went from 20% to 80% of your code (within December)
Brockman, Sequoia AI Ascent · vendor own
Figures: no base statedfigures: unassessedno listed security
Reality anchor, portfolio synthesis. AI Research-Automation Scorecard, code layer Tests whether a code-share claim defines its denominator, can be reproduced and has an independent check.
Evidence and provenance
Figures. not assessed. Coverage 0/0 figure-sentences.
Source relationship. claim relationship unresolved: not assessed.
Snap · 01 Apr 2026
Snap: AI writes more than two-thirds of our new code (credits Anthropic)
Spiegel, Cheeky Pint podcast · vendor ownFortune / Yahoo Finance · trade press
Figures: base partly statedfigures: unassessed
Reality anchor, portfolio synthesis. AI Research-Automation Scorecard, code layer Tests whether a code-share claim defines its denominator, can be reproduced and has an independent check.
Evidence and provenance
Figures. not assessed. Coverage 0/0 figure-sentences.
Source relationship. claim relationship unresolved: not assessed.
Salesforce · 26 Jun 2025
Salesforce: AI is doing 30-50% of the work here now
Benioff, The Circuit · vendor ownCNBC / Bloomberg · trade press
Figures: no base statedfigures: unassessedCRM 193.32 -2.10%
Reality anchor, portfolio synthesis. AI Research-Automation Scorecard, work layer Keeps whole-function and headcount claims separate from code-production claims.
Evidence and provenance
Figures. not assessed. Coverage 0/0 figure-sentences.
Source relationship. claim relationship unresolved: not assessed.
Google DeepMind · 01 May 2025
DeepMind: AlphaEvolve optimised a kernel used to train Gemini (~1% training saving)
DeepMind blog; Novikov · vendor ownVentureBeat / TechBrew · trade press
Figures: base statedfigures: unassessedGOOGL 343.54 -0.08%
Reality anchor, portfolio synthesis. AI Research-Automation Scorecard, research-loop layer Separates an observed tool-assisted saving from autonomous recursive improvement.
Evidence and provenance
Figures. not assessed. Coverage 0/0 figure-sentences.
Source relationship. claim relationship unresolved: not assessed.
Microsoft · 29 Apr 2025
Microsoft: 20-30% of code in our repos is written by software
Nadella, LlamaCon fireside · vendor ownTechCrunch / CNBC · trade press
Figures: base partly statedfigures: unassessedMSFT 492.43 -2.26%
Reality anchor, portfolio synthesis. AI Research-Automation Scorecard, code layer Tests whether a code-share claim defines its denominator, can be reproduced and has an independent check.
Evidence and provenance
Figures. not assessed. Coverage 0/0 figure-sentences.
Source relationship. claim relationship unresolved: not assessed.
Google · 01 Apr 2025
Google: from ~30% to ~75% of new code AI-generated and approved by engineers
Pichai, Q1 2025 earnings call · vendor ownGoogle blog · vendor own
Figures: base partly statedfigures: unassessedGOOGL 343.54 -0.08%
Reality anchor, portfolio synthesis. AI Research-Automation Scorecard, code layer Tests whether a code-share claim defines its denominator, can be reproduced and has an independent check.
Evidence and provenance
Figures. not assessed. Coverage 0/0 figure-sentences.
Source relationship. claim relationship unresolved: not assessed.
Meta · 01 Apr 2025
Meta: maybe half of development done by AI next year (a target)
Zuckerberg, LlamaCon (no current figures) · vendor own
Figures: no base statedfigures: unassessedMETA 578.85 -3.38%
Reality anchor, portfolio synthesis. AI Research-Automation Scorecard, code layer Tests whether a code-share claim defines its denominator, can be reproduced and has an independent check.
Evidence and provenance
Figures. not assessed. Coverage 0/0 figure-sentences.
Source relationship. claim relationship unresolved: not assessed.
Anthropic · 10 Mar 2025
Anthropic: AI writes ~90% of our code, heading to essentially all of it
Amodei, Council on Foreign Relations · vendor ownRedwood Research (re-analysis: nearer 50%) · independent
Figures: no base statedfigures: unassessedno listed security
Reality anchor, portfolio synthesis. AI Research-Automation Scorecard, code layer Tests whether a code-share claim defines its denominator, can be reproduced and has an independent check.
Evidence and provenance
Figures. not assessed. Coverage 0/0 figure-sentences.
Source relationship. claim relationship unresolved: not assessed.
Amazon · 22 Aug 2024
Amazon: 79% of AI code-review suggestions shipped without changes (Amazon Q)
Jassy, LinkedIn/X · vendor ownSlashdot / Yahoo Finance · trade press
Figures: base partly statedfigures: unassessedAMZN 267.28 -1.83%
Reality anchor, portfolio synthesis. AI Research-Automation Scorecard, code layer Tests whether a code-share claim defines its denominator, can be reproduced and has an independent check.
Evidence and provenance
Figures. not assessed. Coverage 0/0 figure-sentences.
Source relationship. claim relationship unresolved: not assessed.
Markets, primary sources, releases and reference panels
S&P 500 7,748.50 +0.26%Nasdaq 26,588.49 +0.54%US 10Y 4.70 -0.42% 08-11VIX 15.28 -1.16% 08-11Brent $93.26 +0.56% 08-11Bitcoin $63,402 -0.20%Nikkei 67,524.06 +0.83%NVDA 224.09 +3.03%MSFT 492.43 -2.26%GOOGL 343.54 -0.08%AMZN 267.28 -1.83%META 578.85 -3.38%ORCL 153.28 +5.36%AMD 482.93 +1.82%AVGO 416.05 -0.01%TSM 429.15 +1.68%MU 911.29 +4.92%CRWV 107.73 +19.28%PLTR 171.04 -2.23%SPCX 146.15 +9.65%ASML 1,810.07 +0.59%STM 54.18 -1.56%SBGSY 71.54 +1.69%BABA 125.22 -2.06%BIDU 104.84 -1.04%TCEHY 56.28 -5.44%SMH 584.83 +2.08%
Freshest data 2026-08-12. A small date after a cell means that cell is older: FRED indices run a trading day behind, and a market closed on a given day has no later price. Each percentage is the move since that series own previous observation, so a gap in the series is a longer window than a day. Context for the announced-vs-delivered lens, not a market call and not investment advice.
Coverage limit: this is a US-listed view of AI hardware, not the market for it
The quote source serves US and OTC ADR listings only, so the reachable China names are platform and cloud companies, not the domestic chipmakers. SMIC, Hua Hong, Cambricon, Samsung and SK Hynix cannot be shown here. OTC ADRs are thinly traded and can lag their home listing.
What that hides: the Shanghai STAR Market listings are where the domestic AI-chip complex is being repriced, and none of them are reachable at this data tier. Moore Threads closed its 5 December 2025 debut up about 425 per cent (offer 114.28 yuan, close 600.50), and MetaX opened up about 693 per cent on 17 December. Neither is a small move and neither can appear above. The absence is a limit of this data tier, not evidence that nothing is happening.
AI Watch: resolution calendar
Dated moments when an open, AI-relevant question resolves, each paired with the specific observable that settles it. These are not predictions. Each row states what to watch and the tell that decides it.
Resolution calendar
2 Aug 2026EU AI Act Article 50 transparency obligations take effect.
Settles on: First enforcement posture and guidance on AI-content transparency and disclosure duties.
late-Sep 2026Is household credit stress turning, or still easing?
Settles on: The Q2 2026 reading of credit-card delinquency at US commercial banks, published around late September because the series runs a quarter behind. Through Q1 2026 it fell every quarter to 2.92 percent, and at small banks outside the top 100 to 6.43 percent, narrowing the gap between them. A rise in either, or a widening gap, would be the turn.
1 Oct 2026Are the Colorado River's 2007 operating rules replaced before they expire?
Settles on: New interim operating guidelines agreed, or a deadlock. A deadlock raises litigation and water-siting risk for Western US data centres.
late-Oct 2026Did the memory squeeze defer enterprise software spending, or end it?
Settles on: IBM's Q3 results, reported through late October. Two numbers decide it: software growth against IBM's own 6 to 8 percent guide, and whether annual recurring revenue keeps growing from the 8 percent it added in Q2. Software printing below 6 percent, or recurring revenue decelerating, would mean the spending was lost rather than delayed.
late-Oct 2026Is the AI data-centre capital-expenditure boom still accelerating?
Settles on: Hyperscaler Q3 earnings, reported through late October: whether 2026 capital expenditure guidance is raised, held or cut, and what is said about high-bandwidth memory allocation for 2027. A held or cut guide would be the first crack. The Q2 round did not produce one, which is why this row was extended rather than closed.
10 Nov 2026Do US controls on lithography exports to China tighten in November?
Settles on: The Affiliates Rule suspension, a carve-out of the current trade truce, expires on 10 November. Three observables: whether it is renewed or allowed to lapse, whether the MATCH Act advances (it would bar all ASML deep-ultraviolet sales to China and start a 150-day clock on Dutch alignment), and what ASML's own guidance says about China revenue.
27 Nov 2026Do China's rare-earth and neon export controls renew, widen or lapse?
Settles on: The renewal decision, with neon controls the specific trigger to watch.
Dec 2026Does data-centre load keep pinning the PJM capacity price at its cap?
Settles on: The 2030/31 base residual auction, the last one under the price collar. Three readings: how much new supply clears against the shortfall, whether the $325 per MW-day cap holds, and the data-centre share of capacity cost. For scale, the 2028/29 auction cleared about 525 MW against a 6,831 MW shortfall, with data centres carrying 40% of capacity cost in that auction.
2 Dec 2026EU AI Act Art. 50(2): does the marking grandfathering bite when it lapses?
Settles on: Whether generative AI systems already on the market before 2 August 2026 carry machine-readable marking and a detection route by the date the transitional period ends.
2 Dec 2027EU AI Act high-risk (Annex III) obligations take effect.
Settles on: The high-risk compliance regime comes into force.
Live gauges
Strait of Hormuz risk (as of 2026-07-31)
Brent peaked at $105.32 on 23 July, held $100.31 on 24 July, then fell to $91.82 by 27 July, the most recent daily observation. The move up ran from a low of $68.53 on 2 July, so the round trip is roughly $37 up and $13 back inside four weeks. Source throughout: FRED series DCOILBRENTEU, Europe Brent spot, which is the same series the board's market strip reads.
Watch: The $100 level flagged here on 13 July was crossed on 23 July and did not hold. The tell moves on again: whether the July low near $68 or the July high near $105 is the better guide to Q4, and whether any renewed disruption puts it back through $100 rather than whether it gets there once.
This still runs against the forecast anchor, but by less than it did. The EIA Short-Term Energy Outlook base case of 7 July had Brent easing to about $70 by Q4 2026 as the tanker backlog cleared, and Rapidan raised its Q4 estimate to near $100 from $85. At $91.82 neither is settled. Goldman's above-$120 case required Hormuz disruption to persist, which on this price path it has not. The base case is recorded as contested, not withdrawn.
Generated from rows explicitly marked for publication in the maintainer's working tracker. The tracker's analysis column is never read by the generator, so private notes cannot reach this file. Public wording is written by hand and carried across regenerations. Nothing is published unless it is marked; the default is private. Status is upcoming until the date passes. The observable, not the outcome, is the point.
Primary sources (16)
Recent papers and public datasets on AI, so a claim can be checked against the underlying research rather than the coverage of it.
PaperDreamFly: Causal Memory and Receding-Horizon Diffusion Planning for Aerial Vision-Language Navigation
arXiv · Yan Deng, Fei Xu · 2026-08-12
PaperAI4AI at Test-Time: Strong-to-Weak Capability Transfer via Harnesses
arXiv · Cheng Qian, Wenting Zhao, Liangwei Yang et al. · 2026-08-12
PaperRedistribution-based Cost Inference Improves Sparse Safe Offline RL
arXiv · Ebenezer Gelo, Geraud Nangue Tasse, Steven James et al. · 2026-08-12
PaperConstructing Dynamic Master Logic Models as Knowledge Graphs for Complex System Diagnostics Using Retrieval-Augmented Large Language Models
arXiv · Saman Marandi, Yu-Shu Hu, Mohammad Modarres · 2026-08-12
PaperA Framework for Designing Reward Functions: From Objectives to Features to Human-Aligned Reward Functions
arXiv · Di Yang Shi, W. Bradley Knox · 2026-08-12
PaperClass Activation Mapping in Explainable Computer Vision: A Method-Centered Review of CNN, Transformer, and Foundation-Model-Era Visual Explanations
arXiv · AmirHossein Eshghi, Hamid Saadatfar, Seyyed Ali Hoseini et al. · 2026-08-12
PaperBeyond Trial-and-Error: Agentic Optimization for Image-to-Video Adherence
arXiv · Aman Tyagi, Hemanth Boinpally, Jonathan Chen et al. · 2026-08-12
PaperVAKRA: Evaluating Multi-Hop Reasoning Across APIs and Retrieval Under Tool-Use Policies
arXiv · Ankita Rajaram Naik, Anupama Murthi, Benjamin Elder et al. · 2026-08-12
Datasetk9cli/video-vec2wav2-tokenizer
Hugging Face · k9cli · 2026-07-22
Datasetanisoleai/fineweb-tokenized
Hugging Face · anisoleai · 2026-05-29
Datasethuggingface/documentation-images
Hugging Face · huggingface · 2026-08-12
DatasetKakologArchives/KakologArchives
Hugging Face · KakologArchives · 2026-08-13
Datasetryanmarten/OpenThoughts-1k-sample
Hugging Face · ryanmarten · 2025-08-31
DatasetSalesforce/wikitext
Hugging Face · Salesforce · 2024-01-04
Datasethf-doc-build/doc-build-dev
Hugging Face · hf-doc-build · 2026-04-20
Datasetayuo/hd_tmp
Hugging Face · ayuo · 2026-07-05
Model releases (81)
81 models in the last 60 days.
2026-08-12 · Bytedance SeedAPI only
ByteDance Seed: Seed 2.1 Turbo
2026-08-12 · Alibaba (Qwen)open weights
Qwen: Qwen3.8 2.4T A95B
2026-08-12 · Bytedance SeedAPI only
ByteDance Seed: Seed-2.0-Code
2026-08-12 · DeepSeekAPI only
DeepSeek: DeepSeek V4 Pro 0813
2026-08-12 · xAIAPI only
SpaceXAI: Grok 4.6
2026-08-11 · Liquidopen weights
LiquidAI: LFM2.5-2.6B (free)
2026-08-11 · Nvidiaopen weights
NVIDIA: Nemotron 3.5 Lightning
2026-08-11 · SakanaAPI only
Sakana: Sakana Namazu
2026-08-10 · UpstageAPI only
Upstage: Solar Pro 4
2026-08-10 · Cohereopen weights
North-Micro-Vision-Instruct
2026-08-09 · Metaopen weights
Meta: Muse Glimmer 30B
2026-08-08 · Alibaba (Qwen)open weights
Qwen3.8-2.4T-A95B-FP8
2026-08-05 · MetaAPI only
Meta: Muse Spark 1.2
2026-08-05 · Nvidiaopen weights
NVIDIA-Nemotron-3.5-Lightning-30B-A3B-Base-BF16
2026-08-05 · Nvidiaopen weights
NVIDIA-Nemotron-3.5-Lightning-30B-A3B-NVFP4-DSpark
2026-08-05 · Nvidiaopen weights
NVIDIA-Nemotron-3.5-Lightning-30B-A3B-NVFP4-DFlash
Sourcing, and what a release list cannot show
Models that reached OpenRouter or the Hugging Face API in the window. Each row says whether the weights are published, which is an attribute of the release rather than a grade: an API-only model is shipped and serving traffic, it is simply not downloadable. Dates are when the model reached the source, not the vendor's own announcement date, and the Hugging Face dates are repo creation, which can sit either side of a public release. The two sides are not counted the same way, so the open and API-only totals are not comparable: OpenRouter is taken in full, while Hugging Face is a fixed list of labs capped at five repositories each. No benchmark scores, by choice. A model announced but never shipped cannot appear in any release list, including this one.
Disclosed conflict: the US government's three hats on AI
A government source commenting on AI holds three positions at once: it regulates the technology, buys it, and holds equity in the stack it comments on. A government AI claim therefore moves three books: policy, procurement and portfolio. Agency primary data (Federal Reserve, BLS, USGS) stays tier 1; this flag applies to government commentary and claims on AI.
Regulator. Gates model releases and sets AI policy (2026 safety-review holds on GPT-5.6 and Fable 5 / Mythos 5).
Customer. Procures AI and awards or withholds federal contracts (Anthropic reportedly frozen out of the stake talks and blocked from federal contracts, mid-2026).
Owner. Holds equity across the AI-buildout stack, and was offered a 5% stake in a model lab (OpenAI).
CFR tracker: roughly $20.9B across 17 equity or quasi-equity deals since Jan 2025 (a broader count reaches 28 firms once letters of intent are included). By dollars: semiconductors ~52%, critical minerals ~43%. Held equity below; revenue-share and proposed positions are kept separate so they are not miscounted.
CompanyGovernment positionStatusStack layer
Intel9.9-10%, $8.9B (Aug 2025) + warrant for 5% moreHeldFab / compute
MP Materials15% (largest shareholder, Jul 2025)HeldRare earths
Lithium Americas5% (some sources 10%) + 5% of the GM Thacker Pass JVHeldCritical minerals
Trilogy Metals10% ($35.6M) + warrants for 7.5% moreHeldCopper / zinc
US Steelgolden share (veto + board seat, no economic stake)HeldGovernance only
Westinghouseup to 20% discussedUnder negotiationNuclear power (DC energy)
IBM~$1B, equity attachedLetter of intentQuantum / AI
GlobalFoundries$375MAnnounced (May 2026)Fab
D-Wave / Rigetti / Infleqtion~$100M eachAnnouncedQuantum
Vulcan Elements$50M equity + Pentagon warrantsAnnouncedRare-earth magnets
Kept separate (not equity):
  • Nvidia: a 25% cut of its China chip sales. Revenue-share, not ownership.
  • OpenAI: 5% offered (~$42.6B on the $852B March-2026 valuation); an early-stage proposal, not held, reportedly with no board seat or vote.
  • Trump personal / brokerage buys (Broadcom, Synopsys, Dell, Intel in Q1) are a separate entity from the federal stakes and are not folded into this roster.
Sources: CNBC, CFR government-deals tracker (via NAI500), Motley Fool, Built In, 24/7 Wall St, TheStreet, CNN, Time, TechCrunch, NOTUS, Crypto Briefing. Verified 2026-07-13.
Conflict of interest: This board's author is assisted by an Anthropic model, and Anthropic is named in this reporting (reportedly frozen out of the stake talks and blocked from federal contracts). Non-neutral on this node; carried as reported.
Method: source class comes from the executable public registry. A numeric claim tier is withheld unless the publisher relationship to the subject resolves. Figure labels state their method and complete-span coverage. Reality anchors require one winning topic rule; tied candidates abstain. Feed selection is editorial and disclosed. This surfaces a structural weakness of a claim; it does not adjudicate truth.

Conflict of interest: the maker is an independent researcher. An Anthropic model helped build the original method and tiers. An OpenAI model later changed the matching, tier and provenance infrastructure. OpenAI is a subject on this board, so that work is a direct conflict and is disclosed rather than treated as an independent check. Anthropic and OpenAI remain subjects under the same published rules. Independent analysis, not investment advice.

Corrections welcome on any judgement here, and they are marked in place with their reason: NMAIResearch@proton.me