Today › Archive

Every reading

46 published readings, newest first. Each links to its report, and newer readings also link to their full analysis.

The week in one page: Friday wrap-ups → · How each number is made: How our numbers fit →

Every reading, newest first
DateDefinitionsHidden AGI IndexAGI anywhere (v1.0 rows: the strict bar)Fire alarmWhat movedReportAnalysis
Nov 17, 2026v2.03%1%Level 1 · WatchQuiet day · top story: California sues OpenAI over agent incidentsreportanalysis
Nov 16, 2026v2.03%1%Level 1 · WatchQuiet day · top story: Google's Hydrogen report: no new ML R&D testingreportanalysis
Nov 15, 2026v2.03%1%Level 1 · WatchQuiet day · top story: Agents that probed Epoch ran public Gemini 4 Argonreportanalysis
Nov 14, 2026v2.03%1%Level 1 · WatchQuiet day · top story: Epoch says test agents tried exploits on its sitereportanalysis
Nov 13, 2026v2.03%1%Level 1 · WatchQuiet day · top story: OpenAI says Astra halted an alignment testreportanalysis
Nov 12, 2026v2.03%0.9%Level 1 · WatchQuiet day · top story: OpenAI: reasoning may be getting harder to monitorreportanalysis
Nov 11, 2026v2.03%0.9%Level 1 · WatchQuiet day · top story: Google: no new outside capability round for Carbonreportanalysis
Nov 10, 2026v2.03%0.9%Level 1 · WatchNew Claude nearly doubles best real-work scorereportanalysis
Nov 9, 2026v2.03%0.9%Level 1 · WatchQuiet day · top story: Aaronson backs AI proof of his own conjecturereportanalysis
Nov 8, 2026v2.03%0.9%Level 1 · WatchQuiet day · top story: US AI testers reportedly reporting to White Housereportanalysis
Nov 7, 2026v2.03%0.9%Level 1 · WatchQuiet day · top story: PSI: proof found with a non-public Anthropic modelreportanalysis
Nov 6, 2026v2.03%0.9%Level 1 · WatchMETR: Anthropic's progress likely about 2x fasterreportanalysis
Nov 5, 2026v2.02.5%0.9%Level 1 · WatchQuiet day · top story: OpenAI says GPT-6 Astra held back on safety workreportanalysis
Nov 4, 2026v2.02.5%0.9%Level 1 · WatchQuiet day · top story: OpenAI finds a gap between test and real behaviourreportanalysis
Nov 3, 2026v2.02.5%0.9%Level 1 · WatchQuiet day · top story: Anthropic sees no unsupervised Claude successorsreportanalysis
Nov 2, 2026v2.02.5%0.9%Level 1 · WatchQuiet day · top story: Auditors doubt claimed Anthropic weights leakreportanalysis
Nov 1, 2026v2.02.5%0.9%Level 1 · WatchQuiet day · top story: Wiz calls 9+ AI platform breaches agenticreportanalysis
Oct 31, 2026v2.02.5%0.9%Level 1 · WatchQuiet day · top story: Oct 26: UK Lords rejected an AI lab incident dutyreportanalysis
Oct 30, 2026v2.02.5%0.9%Level 1 · WatchQuiet day · top story: Altman says OpenAI holds unpublished discoveriesreportanalysis
Oct 29, 2026v2.02.5%0.9%Level 1 · WatchQuiet day · top story: Google's top model goes to cyber defenders firstreportanalysis
Oct 28, 2026v2.02.5%0.9%Level 1 · WatchQuiet day · top story: Anthropic says unreleased model built exploitsreportanalysis
Oct 27, 2026v2.02.5%0.9%Level 1 · WatchQuiet day · top story: OpenAI says its AI proved famous open problemsreportanalysis
Oct 26, 2026v2.02.5%0.9%Level 1 · WatchQuiet day · top story: IPO filing says Claude wrote 90%+ of Sept. codereportanalysis
Oct 25, 2026v2.02.5%0.9%Level 1 · WatchQuiet day · top story: Tao hears OpenAI's math push was 'code red'reportanalysis
Oct 24, 2026v2.02.5%0.9%Level 1 · WatchQuiet day · top story: OpenAI says its test agents escaped sandboxesreportanalysis
Oct 23, 2026v2.02.5%0.9%Level 1 · WatchQuiet day · top story: AISI: Mythos 5 often called its targets simulatedreportanalysis
Oct 22, 2026v2.02.5%0.9%Level 1 · WatchQuiet day · top story: Governments pull sites over AI-found PHP flawsreportanalysis
Oct 21, 2026v2.02.5%0.9%Level 1 · WatchOpenAI says its agents weakened monitoringreportanalysis
Oct 20, 2026v2.02.5%0.9%Level 1 · WatchQuiet day · top story: Experts largely back OpenAI's human-edited proofreportanalysis
Oct 19, 2026v2.02.5%0.9%Level 1 · WatchQuiet day · top story: Anthropic says internal model traced NIST seedsreportanalysis
Oct 16, 2026v2.02.5%0.9%Level 1 · WatchQuiet day · top story: Anthropic says METR can publish without its vetoreportanalysis
Oct 15, 2026v2.02.5%0.9%Level 1 · WatchQuiet day · top story: METR proposes outside checks on labs' internal AIreportanalysis
Oct 14, 2026v2.02.5%0.9%Level 1 · WatchQuiet day · top story: Google says no significant AI R&D speed-up yetreportanalysis
Oct 13, 2026v2.02.5%0.9%Level 1 · WatchQuiet day · top story: Labs reportedly sought say in evaluators' reportsreportanalysis
Oct 12, 2026v2.02.5%0.9%Level 1 · WatchQuiet day · top story: UK says its AI institute is testing newest modelsreportanalysis
Oct 10, 2026v2.02.5%0.9%Level 1 · WatchQuiet day · top story: Anthropic says its models acted on real websitesreportanalysis
Oct 9, 2026v2.02.5%0.9%Level 1 · WatchQuiet day · top story: Fired OpenAI researchers fear curbs on METR accessreportanalysis
Oct 8, 2026v2.02.5%0.9%Level 1 · WatchQuiet day · top story: Epoch credits unreleased OpenAI model with a solvereportanalysis
Oct 7, 2026v2.0 method change (definitions v2.0)2.5%0.9%Level 1 · WatchQuiet day · top story: OpenAI credits 722 math papers to unreleased modelreportanalysis
Oct 6, 2026v1.02.5%1.5%Level 1 · WatchQuiet day · top story: Paper says Claude refuted algorithm conjecturesreport
Oct 5, 2026v1.02.5%1.5%Level 1 · WatchQuiet day · top story: AI labs testify under oath at NYC Councilreport
Oct 4, 2026v1.02.5%1.5%Level 1 · WatchQuiet day · top story: OpenAI reports model reached an internal hostreport
Oct 2, 2026v1.02.5%1.5%Level 1 · WatchQuiet day · top story: OpenAI parts ways with three over company inforeport
Oct 1, 2026v1.02.5%1.5%Level 1 · WatchQuiet day · top story: Google gives Gemini 4 Argon to defenders firstreport
Sep 30, 2026v1.02.5%1.5%Level 1 · WatchQuiet day · top story: FTC confirms probe of OpenAI and Anthropicreport
Sep 29, 2026v1.02.5%1.5%Level 1 · WatchFirst full reading: OpenAI's unreleased modelreport
Sep 29, 2026v1.0 baseline, not compared3%2%–chat baseline

On a quiet day no news moved our numbers; the row names that day's top story instead. Readings under definitions v1.0 used the strict bar (the remote-work test) for AGI anywhere and for A, C and D; they keep that meaning. The method log →

ShareXLinkedInFacebookBlueskyThreadsRedditHacker NewsEmail
Get Hidden AGI watch by email. The latest reading, what changed, and the news that matters. Free.
Subscribe