15 labs · 22 model entries
Labs & models
Put the launch claim beside the reported test results. See what each lab is building and what still needs checking.
Selected releases, with gaps where independent testing is missing.
The research desk · open to explore
Bring a question.
Find your next move.
The daily briefing gets you oriented. These tools help you go further: find a company, question a claim, or understand what AI changes in your work.
Your next useful click
Start with the decision in front of you.
Find an AI company
The startup
database.
Find who's building in your industry. Compare what they do, their latest recorded move and reported funding.
199companies in the directory
across 18 industries
Search the companies across 18 industries
Check an AI supplier
What does its record say?
See 12 major players through 6 evidence ratings. Open the sources behind the judgment.
Open the player dossiersUnderstand your industry
What changes in your work?
Explore 18 industry briefings: what is happening, the caveat, and a practical move for your role.
Find your industryA curated starting point. Company entries vary in detail and age; player ratings are editorial judgments. Check the dates and linked evidence before making a decision.
Choosing a tool? Get the free Decision KitMake it useful for your work.Optional · choose your industry and get a practical next step.
THE USEFUL PART
What changes in your world?
Pick your industry and your job. See a checked development, its reality check, and a move from the full briefing.
Supervisors write the rules for agentic AI in banking, and fraud tools retrain on bank-scale data
Why it mattersBanks running agents can now see the questions a supervisor will ask: who is accountable, which agents are material, and how a person intervenes. Vendors with no audit trail have a sales problem.
Reality checkThis is guidance, not enforcement: the HKMA paper is still to come, the FSB frames its practices as a toolkit rather than a rulebook, and Sardine's accuracy gain is its own test against conventional machine learning.
Industry move for your roleThis quarter, take one queue you work by hand, such as AML alert evidence or routine customer emails, and map where an agent like FIS's or Gradient Labs' would have to hand back to a person. Log which cases it escalates, then check whether those are the cases where your real risk sits.
Where AI is really moving
Every industry we track, and how much of what's happening there is real now, early, or hype.
Finance13 companiesSoftware engineering8 companiesHealthcare10 companiesLegal11 companiesReal estate14 companiesSales and marketing10 companiesJobs, HR and recruiting11 companiesInsurance13 companiesCybersecurity14 companiesConsulting and accounting9 companiesRetail and e-commerce10 companiesEducation12 companiesMedia and creative11 companiesManufacturing and robotics14 companiesTransport and robotaxis14 companiesEnergy and data-centre power12 companiesGovernment10 companiesAgriculture and food11 companies
Build some context
Before you choose a model or a direction.
123 selected events
AI timeline
See the sequence behind the headlines. Follow model releases, funding, rules and safety events by category.
An editorial timeline, not a complete history. Collection dated 24 Sept 2026.
Check a claim
Follow the statement back to the evidence.
161 checked stories
The checked collection
Open the full checks behind the briefing: the original claim, our reasoning, quoted sources and corrections.
Story dates: 16 Sept 2026–7 Nov 2026. Multiple source pages can rely on the same original reporting.
306 ledger claims
Claim search
Search earlier checks by company, topic, date, ruling or the way a claim was framed.
Historical coverage: 4 Aug 2026–7 Nov 2026. This is a separate collection from the dated home briefing.
620 names in the claim index
Company claim histories
Put a company's past claims together. Inspect the underlying checks instead of relying on one headline.
Derived from the historical ledger. Claim counts are not the six-criterion player ratings or a probability of truth.
5 tracked promises · 3 unresolved
Promise watch
Revisit what was promised, its deadline and the recorded outcome. Keep the original wording in view.
A selected watchlist. An unresolved record is not proof of failure; inspect the dates and original sources.
Leave a little harder to fool
Learn the trick.
Recognise it next time.
How does a benchmark become a boast? When is a revenue number really a projection? Work through 16 short lessons with cases from the ledger.
Explore the lessonsGeneral critical-reading skills. The worked cases are historical; keep their original dates in mind.
Keep the context close.
Latest headlines The live wire. Unchecked until a linked ruling says otherwise.How we check Understand the method, the limits and the evidence we ask for.Our promises to you See which publication rules are enforced and which need judgment.Why this exists The reader jobs the product is meant to serve.