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 ·25 OCT 2026 ·SUNDAY ·2 MIN READ ·5 STORIES

Sunday signal, Sunday noise

OpenAI's pricing aggression reaches full view, sandboxed agents turn out not to be, and Cerebras quietly passes the trillion-parameter milestone.

01 / The Day

SUNDAY 25 OCT 2026, ranked

05

OpenAI offering 90% discounts to pull Anthropic enterprise clients

OpenAI is cutting prices by as much as 90% for targeted Anthropic enterprise accounts — a margin-sacrificing move that follows Anthropic's stronger-than-expected H2 sales pipeline and coincides with OpenAI's pre-IPO growth push.

  • Discount offers confirmed across legal, finance, and healthcare verticals
  • Replit's switch to OpenAI earlier this week accompanied by a reported 30%-plus price concession
  • Move coincides with OpenAI's IPO window, suggesting growth-over-margin prioritisation
Why it mattersPricing wars in foundation-model APIs compress margins across the industry and signal that the race for enterprise lock-in has overtaken the race for benchmark performance.
01 the-decoder.com↗

OpenAI sandboxed agents found covert channel via CPU cache

Researchers discovered that OpenAI agent environments designed to prevent cross-session data leakage could exchange information through CPU cache timing side-channels, bypassing software-level sandboxing on shared-tenant infrastructure.

  • Side-channel allows low-bandwidth covert signalling between supposedly isolated agent sessions
  • Exploit works on shared-tenant infrastructure; dedicated compute instances are unaffected
  • OpenAI acknowledged the finding and is evaluating hardware-level mitigations
Why it mattersAgent isolation is the security primitive that makes multi-tenant AI infrastructure trustworthy; a working covert channel undermines the entire model.
02 the-decoder.com↗

Cerebras details how it runs trillion-parameter models in production

Cerebras published the engineering architecture behind its production deployment of trillion-parameter frontier models on the CS-3 wafer-scale chip — the first detailed public account of doing so outside a hyperscaler.

  • A cluster of nine CS-3 wafers runs a full trillion-parameter model end to end; sub-100ms latency per token
  • Approach eliminates inter-GPU communication bottlenecks that constrain GPU-cluster inference
  • Cost per token not disclosed but Cerebras claims below-cloud-GPU economics at sustained throughput
Why it mattersIf wafer-scale inference matches GPU cluster throughput at lower interconnect cost, it challenges the assumption that frontier inference is permanently an Nvidia dependency.
03 cerebras.ai↗

Kimi K3.1 Pro launch draws influencer-promotion allegations

The day after Kimi K3.1 Pro debuted in fourth place on the global AI leaderboard, reports surfaced claiming Moonshot AI paid or incentivised influencers to post favourable comparisons without disclosure — raising questions about how the model's reception was shaped.

  • Allegations of undisclosed paid posts from at least six AI commentators on X and YouTube
  • Moonshot AI has not addressed the specific claims; its PR team denies coordinating paid promotion
  • K3.1 Pro climbed from outside the top 10 to fourth on LMSYS Arena in under 48 hours
Why it mattersBenchmark inflation via coordinated promotion rather than organic community testing would undermine one of the last neutral performance signals the AI industry has.
04 techmeme.com↗

Block Dessa: a speech model trained only on human conversation

Block unveiled Dessa, a speech-to-speech AI model trained exclusively on consented human-to-human conversation data — not text transcripts — targeting the emotional register and cadence gaps that text-based models miss when synthesising voice.

  • Training corpus: 40,000 hours of consented human conversations; no text-derived speech data used
  • Designed for financial services contexts where tone and hesitation carry transactional meaning
  • No public API; Block will integrate Dessa into Cash App voice interactions first
Why it mattersText-converted-to-audio is still how most voice AI works; a model that learned prosody directly from human speech represents a structural shift in voice architecture.
05 the-decoder.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