The register
Everything we are watching, with a state, a date and a written reason for every change. The point of the page is the bottom of it.
- watching 143
- rising 0
- active 0
- fading 0
- retired 0
watching
-
800 V DC rack power
Compute for rent
first seen 2026-09-22
first seen
-
A2A after 1.0
Agents and tooling
first seen 2026-09-22
first seen
-
AGENTS.md as the shared instruction file
Agents and tooling
first seen 2026-09-22
first seen
-
AI Act content marking and labelling
Policy in force
first seen 2026-09-22
first seen
-
AI Act enforcement of general-purpose models
Policy in force
first seen 2026-09-22
first seen
-
AI Act high-risk deferral
Policy in force
first seen 2026-09-22
first seen
-
AI Act in the EEA
Policy in force
first seen 2026-09-22
first seen
-
AMD Helios racks
Compute for rent
first seen 2026-09-22
first seen
-
Accelerators the labs build for themselves
Compute for rent
first seen 2026-09-22
first seen
-
Activation probes as monitors
Research worth knowing
first seen 2026-09-22
first seen
-
Adaptive thinking replacing manual budgets
Models and releases
first seen 2026-09-22
first seen
-
Agent Client Protocol in editors
Agents and tooling
first seen 2026-09-22
first seen
-
Agent Skills as a cross-tool format
Agents and tooling
first seen 2026-09-22
first seen
-
Agent and work-task benchmarks
Research worth knowing
first seen 2026-09-22
first seen
-
Agent frameworks consolidating
Agents and tooling
first seen 2026-09-22
first seen
-
Agent identity in directories and OAuth
Agents and tooling
first seen 2026-09-22
first seen
-
Agent payment and checkout protocols competing
Agents and tooling
first seen 2026-09-22
first seen
-
Agentic misalignment evaluations
Research worth knowing
first seen 2026-09-22
first seen
-
Agents reviewing and approving pull requests
Agents and tooling
first seen 2026-09-22 · 1 note
first seen
-
Assistant and coding plan ladders
Agents and tooling
first seen 2026-09-22
first seen
-
Audio billed per minute beside tokens
Models and releases
first seen 2026-09-22
first seen
-
Audit access to agent sessions
Agents and tooling
first seen 2026-09-22
first seen
-
Automated research agents
Research worth knowing
first seen 2026-09-22
first seen
-
Benchmark re-issues
Research worth knowing
first seen 2026-09-22
first seen
-
Bills of materials for models and datasets
Agents and tooling
first seen 2026-09-22
first seen
-
Cache-read prices diverging by model
Models and releases
first seen 2026-09-22
first seen
-
Chain-of-thought monitorability
Research worth knowing
first seen 2026-09-22
first seen
-
Chip export licensing
Policy in force
first seen 2026-09-22
first seen
-
Cloudflare's Accountable crawler designation
Agents and tooling
first seen 2026-09-22
first seen
-
Co-packaged optics switches
Compute for rent
first seen 2026-09-22
first seen
-
Coding benchmark saturation
Research worth knowing
first seen 2026-09-22
first seen
-
Coding tools changing owner
Agents and tooling
first seen 2026-09-22
first seen
-
Coding tools retiring models on their own clock
Models and releases
first seen 2026-09-22
first seen
-
Companion chatbot rules
Policy in force
first seen 2026-09-22
first seen
-
Comprehensive state AI acts
Policy in force
first seen 2026-09-22
first seen
-
Compute in Europe
Compute for rent
first seen 2026-09-22
first seen
-
Content Credentials in capture and display
Agents and tooling
first seen 2026-09-22
first seen
-
Cyber Resilience Act reporting
Policy in force
first seen 2026-09-22
first seen
-
Dateless pinned model ids
Models and releases
first seen 2026-09-22
first seen
-
Deepfake takedown duties
Policy in force
first seen 2026-09-22
first seen
-
Desk-sized inference machines
Compute for rent
first seen 2026-09-22
first seen
-
Diffusion language models
Research worth knowing
first seen 2026-09-22
first seen
-
Embedding models retired for successors
Models and releases
first seen 2026-09-22
first seen
-
Energy per prompt disclosures
Compute for rent
first seen 2026-09-22
first seen
-
Evaluation awareness and sandbagging
Research worth knowing
first seen 2026-09-22
first seen
-
Fair use for training
Policy in force
first seen 2026-09-22
first seen
-
Federal pre-emption of state AI laws
Policy in force
first seen 2026-09-22
first seen
-
Fine-tuning job creation restricted
Models and releases
first seen 2026-09-22
first seen
-
Foundation-model medical devices
Policy in force
first seen 2026-09-22
first seen
-
Four-bit inference formats
Compute for rent
first seen 2026-09-22
first seen
-
Frontier model transparency laws
Policy in force
first seen 2026-09-22
first seen
-
Frontier models behind gated programmes
Models and releases
first seen 2026-09-22
first seen
-
Frontier risk disclosures
Policy in force
first seen 2026-09-22
first seen
-
Gateways between agents and the model API
Agents and tooling
first seen 2026-09-22
first seen
-
Genome models at base resolution
Research worth knowing
first seen 2026-09-22
first seen
-
Gigawatt campuses coming online
Compute for rent
first seen 2026-09-22
first seen
-
Google's coding agents consolidating
Agents and tooling
first seen 2026-09-22
first seen
-
Government model evaluations
Policy in force
first seen 2026-09-22
first seen
-
HBM4 supply
Compute for rent
first seen 2026-09-22
first seen
-
Harmonised standards for the AI Act
Policy in force
first seen 2026-09-22
first seen
-
Hosted runtimes that run an agent's sessions
Agents and tooling
first seen 2026-09-22
first seen
-
Hybrid and linear attention in shipped models
Research worth knowing
first seen 2026-09-22
first seen
-
Image models retired for successors
Models and releases
first seen 2026-09-22
first seen
-
Inference providers selling capacity
Compute for rent
first seen 2026-09-22
first seen
-
Inference-only accelerators
Compute for rent
first seen 2026-09-22
first seen
-
Interactive benchmarks
Research worth knowing
first seen 2026-09-22
first seen
-
Ironwood TPUs for rent
Compute for rent
first seen 2026-09-22
first seen
-
Labs committing to hyperscaler compute
Compute for rent
first seen 2026-09-22
first seen
-
Learned weather models in operational use
Research worth knowing
first seen 2026-09-22
first seen
-
Long-context price tiers
Models and releases
first seen 2026-09-22 · 1 note
first seen
-
MCP SDK tiering
Agents and tooling
first seen 2026-09-22
first seen
-
MCP's 2026-07-28 revision and its deprecation clock
Agents and tooling
first seen 2026-09-22
first seen
-
Machine-checked mathematics
Research worth knowing
first seen 2026-09-22
first seen
-
Machine-readable AI usage preferences
Agents and tooling
first seen 2026-09-22
first seen
-
Machine-readable marking of generated content
Agents and tooling
first seen 2026-09-22
first seen
-
Maximum output limits rising
Models and releases
first seen 2026-09-22 · 1 note
first seen
-
Meta's lineup moving from downloadable Llama to served Muse
Models and releases
first seen 2026-09-22
first seen
-
Muon and second-order optimisers
Research worth knowing
first seen 2026-09-22
first seen
-
NIST AI RMF revision
Policy in force
first seen 2026-09-22
first seen
-
NVLink Fusion partners
Compute for rent
first seen 2026-09-22
first seen
-
Neocloud power online
Compute for rent
first seen 2026-09-22
first seen
-
Neural accelerators in Apple GPUs
Compute for rent
first seen 2026-09-22
first seen
-
Official MCP extensions reaching hosts
Agents and tooling
first seen 2026-09-22
first seen
-
Older models closed to new users
Models and releases
first seen 2026-09-22
first seen
-
On-policy distillation
Research worth knowing
first seen 2026-09-22
first seen
-
One-million-token windows as the default
Models and releases
first seen 2026-09-22 · 2 notes
first seen
-
Open video datasets
Research worth knowing
first seen 2026-09-22
first seen
-
Open weights at frontier scale
Models and releases
first seen 2026-09-22
first seen
-
Openly licensed pretraining corpora
Research worth knowing
first seen 2026-09-22
first seen
-
Operators disclosing agents acting outside sanction
Research worth knowing
first seen 2026-09-22
first seen
-
Paying for crawled content
Agents and tooling
first seen 2026-09-22
first seen
-
Peer review under volume
Research worth knowing
first seen 2026-09-22
first seen
-
Per-conversation and per-action pricing for agents
Agents and tooling
first seen 2026-09-22
first seen
-
Permission modes that let agents act without a prompt
Agents and tooling
first seen 2026-09-22
first seen
-
Permissive licences on the largest open weights
Models and releases
first seen 2026-09-22
first seen
-
Plans moving from seats to usage
Agents and tooling
first seen 2026-09-22
first seen
-
Plugin marketplaces for coding agents
Agents and tooling
first seen 2026-09-22
first seen
-
Price changes announced with a date
Models and releases
first seen 2026-09-22
first seen
-
Priority and fast tiers priced as multipliers
Models and releases
first seen 2026-09-22
first seen
-
Product liability for software
Policy in force
first seen 2026-09-22
first seen
-
Publisher and platform licensing agreements
Policy in force
first seen 2026-09-22
first seen
-
ROCm as a serving target
Compute for rent
first seen 2026-09-22
first seen
-
Regional inference at a premium
Models and releases
first seen 2026-09-22
first seen
-
Residual stream redesigns
Research worth knowing
first seen 2026-09-22
first seen
-
Retirement notice periods
Models and releases
first seen 2026-09-22
first seen
-
Retroactive opt-out in training data
Research worth knowing
first seen 2026-09-22
first seen
-
Reward hacking and emergent misalignment
Research worth knowing
first seen 2026-09-22
first seen
-
Robot policies released with weights or an API
Models and releases
first seen 2026-09-22
first seen
-
Safeguard changes on models already served
Models and releases
first seen 2026-09-22
first seen
-
Safeguards for open-weight release
Research worth knowing
first seen 2026-09-22
first seen
-
Safety framework revisions
Policy in force
first seen 2026-09-22
first seen
-
Same model, different outlet, different clock
Models and releases
first seen 2026-09-22
first seen
-
Sampling parameters withdrawn from APIs
Models and releases
first seen 2026-09-22
first seen
-
Search and retrieval layers sold to agents
Agents and tooling
first seen 2026-09-22
first seen
-
Serious incident reporting
Policy in force
first seen 2026-09-22
first seen
-
Shadow-library training data
Policy in force
first seen 2026-09-22
first seen
-
Signed requests from agents on the open web
Agents and tooling
first seen 2026-09-22
first seen
-
Sparse attention for million-token context
Research worth knowing
first seen 2026-09-22
first seen
-
Sparse autoencoders and transcoders
Research worth knowing
first seen 2026-09-22
first seen
-
Speech models retired for successors
Models and releases
first seen 2026-09-22
first seen
-
Synthetic data in pretraining corpora
Research worth knowing
first seen 2026-09-22
first seen
-
Task time horizons
Research worth knowing
first seen 2026-09-22
first seen
-
Task-specific variants of a frontier family
Models and releases
first seen 2026-09-22
first seen
-
Test-time compute and overthinking
Research worth knowing
first seen 2026-09-22
first seen
-
Tier and codename proliferation inside families
Models and releases
first seen 2026-09-22
first seen
-
Tokenisers changing what a token buys
Models and releases
first seen 2026-09-22
first seen
-
Tools choosing the model per task
Agents and tooling
first seen 2026-09-22 · 1 note
first seen
-
Training content summaries shipped with releases
Models and releases
first seen 2026-09-22
first seen
-
UALink silicon
Compute for rent
first seen 2026-09-22
first seen
-
UK AI legislation
Policy in force
first seen 2026-09-22
first seen
-
UK text and data mining exception
Policy in force
first seen 2026-09-22
first seen
-
Ultra Ethernet fabrics
Compute for rent
first seen 2026-09-22
first seen
-
Unsolved-problem benchmarks
Research worth knowing
first seen 2026-09-22
first seen
-
Vendors retiring hosted evaluation platforms
Agents and tooling
first seen 2026-09-22
first seen
-
Vera Rubin systems
Compute for rent
first seen 2026-09-22
first seen
-
Verifier-gated supervision
Research worth knowing
first seen 2026-09-22
first seen
-
Video APIs withdrawn
Models and releases
first seen 2026-09-22
first seen
-
Voice in coding agents
Agents and tooling
first seen 2026-09-22
first seen
-
WebMCP as a browser standard
Agents and tooling
first seen 2026-09-22
first seen
-
Weight storage and formats on Hugging Face
Agents and tooling
first seen 2026-09-22
first seen
-
Whether consumer and API data train the vendor's models by default
Models and releases
first seen 2026-09-22
first seen
-
Worktree isolation in coding agents
Agents and tooling
first seen 2026-09-22
first seen
-
llms.txt files on the open web
Agents and tooling
first seen 2026-09-22
first seen
How a topic moves
watching → rising → active → fading → retired. A change is proposed with numbers, decided by an editor, and the reason goes on the record. Nothing moves automatically, because automatic states flap: a name that spikes for one conference is not rising. A name that has gone quiet because it is finished and widely used is not fading.