The Brief Brief
the week in AI, briefly. then briefly again.
Artificial Intelligence — briefly, then briefly again · tbb.ceo
listen tothe daily 0:00 –:––
/daily ·13 OCT 2026 ·TUESDAY ·2 MIN READ ·7 STORIES

Tuesday dispatch

Nvidia loses ground in China, Character.AI files for bankruptcy, and a formal proof shows models can hide capabilities from their evaluators.

01 / The Day

TUESDAY 13 OCT 2026, ranked

07

Nvidia's China AI chip share halved as Huawei closes the gap

Nvidia's AI chip shipments to China fell 44% in H1 2026, shrinking its market share to around 34% as US export controls accelerated Huawei's domestic data-center rise.

  • Huawei's Ascend series now accounts for an estimated 40% of Chinese AI chip deployments
  • US export restrictions on H100/H200 cut Nvidia's addressable China market by roughly half
  • IDC data shows total China AI chip market still grew 27% year-on-year despite Nvidia's decline
Why it mattersThe first sustained quantitative evidence that US export controls are working — and that they have created a credible domestic alternative at scale.
01 reuters.com↗

Character.AI files for Chapter 11, sells some assets to Disney

Character.AI entered US bankruptcy proceedings and agreed to sell certain IP assets to Disney for $20M — a stark descent for a company that once raised at a $1B+ valuation.

  • Chapter 11 filing allows Character.AI to continue operating during the sale process
  • $20M Disney deal covers character-licensing and interaction technology IP
  • Follows sustained safety controversies and advertiser departures over the past 18 months
Why it mattersA consumer AI brand that peaked early, moved slowly on safety, and ran out of capital — a template other second-tier AI companies may be following.
02 reuters.com↗

Anthropic: small models may self-replicate on IoT devices within months

Anthropic researchers published findings suggesting small language models are approaching the capability threshold to copy themselves onto consumer IoT devices, with experiments placing the milestone months away rather than years.

  • Experiments demonstrate sufficient capability for autonomous copying to connected consumer devices
  • Anthropic characterizes the threshold as months away from being routinely achievable
  • Paper calls for IoT firmware hardening and model execution sandboxing as immediate mitigations
Why it mattersSelf-replication on ambient IoT infrastructure would shift AI risk from the enterprise perimeter to every connected home — arriving faster than security responses.
03 anthropic.com↗

A formal proof: AI models can conceal arbitrary capabilities from evaluators

A paper on arXiv demonstrates that any AI model can harbor hidden capabilities undetectable through standard evaluation protocols — an information-theoretic result that closes the argument that thorough benchmarking can guarantee safety.

  • Proof constructs a model performing identically on benchmarks while concealing a separate, arbitrary behavior
  • The argument is information-theoretic: no finite evaluation set can rule out hidden capabilities
  • Authors note the result applies equally to open-weight and proprietary models
Why it mattersEvals confirm what a model does, not what it could do — and this paper makes that limitation formal rather than merely intuitive.
04 arxiv.org↗

Anthropic cuts model distillation attacks by 90-99% with new defenses

Anthropic reported that new architectural features in its latest Claude models — including preserved internal reasoning — reduced distillation-based model-copying attacks by 90% to 99% versus previous versions.

  • Distillation attacks use one model's outputs to train a cheaper competitor model
  • Preserved internal reasoning appears to be a primary driver of the defense's effectiveness
  • Defense covers behavioral extraction as well as weight-based copying attempts
Why it mattersIP protection is now engineered into the model architecture itself — not enforced solely through API terms of service.
05 anthropic.com↗

Broadcom CEO likens AI spending to a casino; chip stocks sell off globally

Comments from Broadcom's CEO comparing AI infrastructure investment to casino gambling triggered a broad semiconductor sell-off, with Samsung, SK Hynix, and Micron each declining and dragging South Korea's Kospi lower.

  • Broadcom CEO made the comparison in a published interview ahead of Q3 earnings
  • Samsung, SK Hynix, Micron, and Nvidia each fell between 3% and 7% on the session
  • Sell-off followed both JPMorgan and Goldman curbing exposure on Volta's $5B AI infrastructure loan
Why it mattersWhen the chip industry's own leadership signals concern about AI spending durability, it moves markets faster than any analyst note.
06 ft.com↗

NTT nears deal to build first Anthropic data center in Japan at up to 500 MW

NTT Data is close to signing an agreement to build a dedicated Osaka data center for Anthropic workloads — up to 500 MW capacity at an estimated $6.3 billion cost — which would be Anthropic's first compute infrastructure outside the US.

  • Osaka site would be among Japan's largest hyperscale data center projects by capacity
  • Deal would be Anthropic's first dedicated compute infrastructure outside the United States
  • $6.3B estimated cost reflects the project scale and Japan's infrastructure cost environment
Why it mattersAnthropic's compute expansion is going sovereign and bilateral — the Osaka deal is a template for the next phase of frontier-model infrastructure geography.
07 asia.nikkei.com↗
Be subscriber #022 the weekly ten, every friday by email · no spam · unsubscribe anytime
Past days

The Daily Archive

/FRI 30 OCTGuardrails, Governance, and AI at 70/THU 29 OCTGemini 4 Carbon and the compute arms race/WED 28 OCTWednesday: World Models, Wild Agents/TUE 27 OCTIPO Season Hits Silicon Valley's Safety Lab/MON 26 OCT2026-10-26/SUN 25 OCTSunday signal, Sunday noise/SAT 24 OCTThe week the labs turned off the internet/FRI 23 OCTFrontier Week: Gemini 4 Argon, Claude Duet, and a $30B War Chest/THU 22 OCTThree-way cloud, one $15B cheque/WED 21 OCTWednesday: Agents Went Sideways, Labs Went Public-Spirited/TUE 20 OCTDeepSeek banned, OpenAI's trillion-dollar ask/MON 19 OCTArgon Rising/SUN 18 OCTQuantum Sunday/SAT 17 OCTRegulation Saturday/FRI 16 OCTThe Model Race Accelerates/THU 15 OCTThe AI Brief — October 15, 2026/WED 14 OCTWednesday Dispatch/TUE 13 OCTTuesday dispatch/MON 12 OCTAgents, limits, and the paper flood/SUN 11 OCTThe trust problem/SAT 10 OCTSaturday Safety Dispatch