Today
Thursday, 12 November 2026
The morning's sourced briefing on AI, technology and enterprise risk: the developments that matter most, what each one means, and which forecasts it moves.
Listen
Today's briefing, read aloud from start to finish.
Why it matters
The big picture
Today showed intent without implementation in chip policy and results that depend on the test in AI measurement, and both leave decision makers planning under uncertainty.
Read the reasoning
US export policy on Nvidia chips has been authorised by the president three times this year and still stalls inside Commerce, while senators from both parties and a leakage estimate of 1.6 million H100-equivalents raise the cost of inaction. For firms exposed to China-linked supply chains, the risk is a policy that can tighten abruptly. Compliance preparation should proceed before the formal rule appears.
On the capability side, ARC Prize and Epoch both report new highs for Anthropic and Google models, but the ARC results differ by up to 34 points depending on harness. A headline number is no longer a reliable basis for deployment decisions, and buyers need results that match their own conditions of cost, runtime and tooling.
In Europe, the Commission continues to build the machinery of the AI Act through binding sandbox rules, while the DSA election guidelines consultation, not covered in today's edition, signals that generative AI is being folded into platform risk duties. Regulation is moving by procedure, and firms that track it closely will be better placed than those that wait for headline deadlines.
- Commerce enforcement of chip diversion
Watch whether presidential approval becomes a formal Commerce action, such as new listings or rules on resale and remote access.
- Harness-dependent benchmark reporting
Gaps between provider-adapter and standard harness results are pushing buyers toward harness-matched, cost-aware evaluation of agentic models.
- EU AI Act operational detail
Implementing acts such as the sandbox rules show the AI Act moving from text to supervised practice across member states.
Developments
What moved
- 01Export ControlsUS Department of Commerce · United States / China
Axios: Trump again backs curbs on Nvidia chip diversion to China while Lutnick stalls at Commerce
Why it matters. US chip enforcement is a policy that is announced repeatedly but not implemented, and buyers cannot plan around it. Cloud providers, data centre operators and AI hardware buyers with China-linked customers or supply chains are exposed to a sudden tightening or to continued drift. They should map resale, intermediary and remote-access exposure now and watch for any Commerce action that turns presidential approval into a formal measure.
“When a rule is announced again and again but never carried out, the people who depend on it are left guessing. Clear and lasting rules let everyone plan ahead.”
Jorai Laurenceo
Founder & Executive Director, LaurenceoWhat happened and sources
What happened Sourced reporting
Axios reports that President Trump authorised a crackdown on companies suspected of diverting Nvidia's advanced AI chips to China for the third time this year, and consulted Elon Musk before doing so. Sources say Commerce Secretary Howard Lutnick has resisted efforts by other administration officials to restrict companies suspected of diverting the chips. Lutnick reportedly warned that the controls would start a trade war with China, while Musk backed the case made by Rubio, Bessent, Greer and Hegseth.
- AxiosScoop: Lutnick stalled Trump's Nvidia–China chip curbs after Oval talks
- 02Export ControlsUS Congress / Epoch AI · United States / China
Warren and Banks urge Commerce to close chip loopholes; Epoch estimates 1.6m H100-equivalents reached China
Why it matters. Cross-party pressure plus a quantified leakage figure makes tighter rules on chip resale and access more likely, whatever the internal dispute at Commerce. Cloud and data centre operators in transit hubs, especially in Southeast Asia, carry the greatest exposure to new due-diligence duties. They should tighten customer verification and end-use documentation before regulators require it.
What happened and sources
What happened Sourced reporting
Al Jazeera reports that US lawmakers have invoked the Nvidia controversy to push for tighter limits, with Senators Warren and Banks urging Commerce to close export control loopholes so advanced AI chips do not reach China's military. It cites an Epoch AI estimate that chips with compute equivalent to 1.6 million Nvidia H100s reached China in violation of export restrictions up to the end of 2025. Authorities and analysts have documented widespread smuggling of restricted chips into China, mainly through Southeast Asia.
- Al JazeeraNvidia spy scandal fuels debate over chip exports to China
- 03AI BenchmarksARC Prize Foundation · Global
ARC Prize verifies Fable 5.2 and Gemini 4 Carbon scores on ARC-AGI-3, 1 to 3 benchmarks
Why it matters. Headline scores now depend on the test harness more than on the model, with gaps of 16 and 34 points between harnesses. Enterprises choosing models for agentic work should demand harness-matched results and cost and runtime data rather than rely on the best published figure. The runtime-limit finding points to latency and compute cost as practical constraints even where accuracy is high.
What happened and sources
What happened Sourced reporting
ARC Prize verified Claude Fable 5.2 at 97.7% on ARC-AGI-3 with the provider adapter harness and 81.2% with the standard harness, and said this is a new high in the standard harness. At max reasoning with the standard harness, 28 of 55 semi-private games reached the 12-hour runtime limit. For Google's Gemini 4 Carbon it verified 99.1% with the provider adapter and 65.1% with the standard harness, and said Carbon beat its predecessor Argon by 18 points on ARC-AGI-3 in the standard harness.
- ARC Prize (X)Claude Fable 5.2 from @AnthropicAI on ARC-AGI (Verified)
- ARC Prize (X)Gemini 4 Carbon from @Google on ARC-AGI (Verified)
- 04Frontier AIEpoch AI · Global
Epoch AI reports Claude Fable 5.2 sets a new Epoch Capabilities Index record in pre-launch testing
Why it matters. An independent composite index gives technology buyers a firmer read on the frontier than vendor-reported scores, and it now places Anthropic at the top. Procurement and strategy teams should treat the figure as provisional and wait for Epoch's full results before shifting model commitments. The margin over the previous leader is what to watch once the index settles.
What happened and sources
What happened Sourced reporting
Epoch AI's benchmarks hub, dated 9 November 2026, says Claude Fable 5.2 set a new record on the Epoch Capabilities Index. Epoch says the result rests on pre-launch testing using only its own benchmarks and may change as more results become available.
- Epoch AIAI benchmarks & capabilities
- 05AI RegulationEuropean Commission · European Union
EU publishes Implementing Regulation 2026/2487 setting common rules for AI regulatory sandboxes
Why it matters. Uniform sandbox rules give AI developers and regulated firms a more predictable route to supervised testing across member states. Financial services, health and other regulated sectors planning EU-facing AI deployments should review the regulation before choosing where to test. National authorities now need to set up sandboxes to a common standard, and the quality of that execution is the next thing to watch.
What happened and sources
What happened Sourced reporting
Commission Implementing Regulation (EU) 2026/2487 was published in the Official Journal on 11 November 2026. Adopted under Article 58(1) of the AI Act, it sets detailed rules for establishing, operating and supervising AI regulatory sandboxes. The Commission considers common rules necessary to avoid fragmentation and ensure consistent implementation of the AI Act.
- EU Law LiveCommission lays down common rules for AI regulatory sandboxes
Analysis
Laurenceo's take
The daily View on the lead story, the latest forecast move and the newest long reads. All analysis
Sources
Where the evidence comes from
Every development is traced to the source that reported it. Candidates that cannot be verified are rejected before anything is written.
Archive
Past briefings
Each briefing is stored on its own and never overwritten. Corrections are marked on the edition and listed in the corrections log.