ainotis Join
My notis

On the move

The topics we follow, and how often they appear in our stories each week.

143
Watching
0
Rising
0
Active
0
Fading
0
Retired

Every topic starts at Watching. Three weeks of counts are not enough yet for any to move; the registrar proposes a move, our editor decides.

Picking up

At least three stories from two sources in four weeks.

Laws and rulingsFrontier risk disclosures 7 stories · 14 sources Money and computeLabs committing to hyperscaler compute 6 stories · 101 sources Models and releasesFrontier models behind gated programmes 5 stories · 103 sources ResearchOperators disclosing agents acting outside sanction 5 stories · 100 sources Money and computeVera Rubin systems 5 stories · 11 sources Money and computeAccelerators the labs build for themselves 3 stories · 4 sources

Watching · 143

most mentioned first
Frontier risk disclosuresLaws and rulingsLabs committing to hyperscaler computeMoney and computeFrontier models behind gated programmesModels and releasesOperators disclosing agents acting outside sanctionResearchVera Rubin systemsMoney and computeAccelerators the labs build for themselvesMoney and computeAgent identity in directories and OAuthAgents and appsAgent payment and checkout protocols competingAgents and appsOpen weights at frontier scaleModels and releasesSafety framework revisionsLaws and rulingsTools choosing the model per taskAgents and appsAgent and work-task benchmarksResearch
Show all 143Agent Skills as a cross-tool formatAgents and appsAgents reviewing and approving pull requestsAgents and appsAssistant and coding plan laddersAgents and appsChip export licensingLaws and rulingsCompute in EuropeMoney and computeFair use for trainingLaws and rulingsGateways between agents and the model APIAgents and appsGigawatt campuses coming onlineMoney and computeGovernment model evaluationsLaws and rulingsHosted runtimes that run an agent's sessionsAgents and appsOne-million-token windows as the defaultModels and releasesSafeguards for open-weight releaseResearchSame model, different outlet, different clockModels and releasesAgentic misalignment evaluationsResearchAI Act content marking and labellingLaws and rulingsAMD Helios racksMoney and computeAudio billed per minute beside tokensModels and releasesAutomated research agentsResearchChain-of-thought monitorabilityResearchCoding tools changing ownerAgents and appsEvaluation awareness and sandbaggingResearchInference providers selling capacityMoney and computeInference-only acceleratorsMoney and computeLong-context price tiersModels and releasesMachine-checked mathematicsResearchMachine-readable marking of generated contentAgents and appsNeocloud power onlineMoney and computePermissive licences on the largest open weightsModels and releasesPrice changes announced with a dateModels and releasesPriority and fast tiers priced as multipliersModels and releasesProduct liability for softwareLaws and rulingsSafeguard changes on models already servedModels and releasesSearch and retrieval layers sold to agentsAgents and appsTask time horizonsResearchTask-specific variants of a frontier familyModels and releasesWhether consumer and API data train the vendor's models by defaultModels and releases800 V DC rack powerMoney and computeA2A after 1.0Agents and appsActivation probes as monitorsResearchAdaptive thinking replacing manual budgetsModels and releasesAgent Client Protocol in editorsAgents and appsAgent frameworks consolidatingAgents and appsAGENTS.md as the shared instruction fileAgents and appsAI Act enforcement of general-purpose modelsLaws and rulingsAI Act high-risk deferralLaws and rulingsAI Act in the EEALaws and rulingsAudit access to agent sessionsAgents and appsBenchmark re-issuesResearchBills of materials for models and datasetsAgents and appsCache-read prices diverging by modelModels and releasesCloudflare's Accountable crawler designationAgents and appsCo-packaged optics switchesMoney and computeCoding benchmark saturationResearchCoding tools retiring models on their own clockModels and releasesCompanion chatbot rulesLaws and rulingsComprehensive state AI actsLaws and rulingsContent Credentials in capture and displayAgents and appsCyber Resilience Act reportingLaws and rulingsDateless pinned model idsModels and releasesDeepfake takedown dutiesLaws and rulingsDesk-sized inference machinesMoney and computeDiffusion language modelsResearchEmbedding models retired for successorsModels and releasesEnergy per prompt disclosuresMoney and computeFederal pre-emption of state AI lawsLaws and rulingsFine-tuning job creation restrictedModels and releasesFoundation-model medical devicesLaws and rulingsFour-bit inference formatsMoney and computeFrontier model transparency lawsLaws and rulingsGenome models at base resolutionResearchGoogle's coding agents consolidatingAgents and appsHarmonised standards for the AI ActLaws and rulingsHBM4 supplyMoney and computeHybrid and linear attention in shipped modelsResearchImage models retired for successorsModels and releasesInteractive benchmarksResearchIronwood TPUs for rentMoney and computeLearned weather models in operational useResearchllms.txt files on the open webAgents and appsMachine-readable AI usage preferencesAgents and appsMaximum output limits risingModels and releasesMCP SDK tieringAgents and appsMCP's 2026-07-28 revision and its deprecation clockAgents and appsMeta's lineup moving from downloadable Llama to served MuseModels and releasesMuon and second-order optimisersResearchNeural accelerators in Apple GPUsMoney and computeNIST AI RMF revisionLaws and rulingsNVLink Fusion partnersMoney and computeOfficial MCP extensions reaching hostsAgents and appsOlder models closed to new usersModels and releasesOn-policy distillationResearchOpen video datasetsResearchOpenly licensed pretraining corporaResearchPaying for crawled contentAgents and appsPeer review under volumeResearchPer-conversation and per-action pricing for agentsAgents and appsPermission modes that let agents act without a promptAgents and appsPlans moving from seats to usageAgents and appsPlugin marketplaces for coding agentsAgents and appsPublisher and platform licensing agreementsLaws and rulingsRegional inference at a premiumModels and releasesResidual stream redesignsResearchRetirement notice periodsModels and releasesRetroactive opt-out in training dataResearchReward hacking and emergent misalignmentResearchRobot policies released with weights or an APIModels and releasesROCm as a serving targetMoney and computeSampling parameters withdrawn from APIsModels and releasesSerious incident reportingLaws and rulingsShadow-library training dataLaws and rulingsSigned requests from agents on the open webAgents and appsSparse attention for million-token contextResearchSparse autoencoders and transcodersResearchSpeech models retired for successorsModels and releasesSynthetic data in pretraining corporaResearchTest-time compute and overthinkingResearchTier and codename proliferation inside familiesModels and releasesTokenisers changing what a token buysModels and releasesTraining content summaries shipped with releasesModels and releasesUALink siliconMoney and computeUK AI legislationLaws and rulingsUK text and data mining exceptionLaws and rulingsUltra Ethernet fabricsMoney and computeUnsolved-problem benchmarksResearchVendors retiring hosted evaluation platformsAgents and appsVerifier-gated supervisionResearchVideo APIs withdrawnModels and releasesVoice in coding agentsAgents and appsWebMCP as a browser standardAgents and appsWeight storage and formats on Hugging FaceAgents and appsWorktree isolation in coding agentsAgents and apps

Topics

The topics we follow, and how often they appear in our stories each week.

Picking up

At least three stories from two sources in four weeks.

Watching 143

Rising 0

None yet.

Active 0

None yet.

Fading 0

None yet.

Retired 0

None yet.

Counts cover only the stories we published, not all AI news. Small numbers can mislead.

Every change, with the reason

How a topic changes status

watching → rising → active → fading → retired. An editor decides every change and records the reason. Nothing changes automatically.