the week in AI, briefly. then briefly again.
Artificial Intelligence — briefly, then briefly again · tbb.ceo
listen tothe daily 0:00 –:––
/daily ·18 NOV 2026 ·WEDNESDAY ·2 MIN READ ·7 STORIES

New models, an IPO, and the first AI cyberattack loop

OpenAI launches GPT-6.2, Anthropic begins trading on Nasdaq, and Google documents the first fully autonomous AI-driven attack chain — all in 24 hours.

01 / The Day

WEDNESDAY 18 NOV 2026, ranked

07

Google catches first end-to-end autonomous AI cyberattack pipeline

Google's Threat Intelligence Group documented exploit_agent, a multi-agent framework that autonomously monitored patch releases, reverse-engineered vulnerabilities, synthesised working exploits, and deployed webshells — with no human in the loop at any stage.

  • Zero-human pipeline: patch monitor → differential bug analysis → exploit synthesis → webshell deployment
  • Agents hit public-facing targets within hours of CVE disclosure
  • Google sharing detection signatures with CISA and major cloud operators
Why it mattersThe first fully autonomous end-to-end attack pipeline means defenders must operate at machine speed — the dwell-time assumptions in most incident response playbooks are now wrong.

OpenAI launches GPT-6.2 Astra and Sol — API price down 20%

OpenAI released two new models: GPT-6.2 Astra, its frontier reasoning tier with new highs on Terminal-Bench-Science and Agents' Last Exam, and GPT-6.2 Sol for lower-latency everyday tasks. Astra API input pricing drops 20% to $8 per million tokens.

  • Astra: new state of the art on Terminal-Bench-Science and Agents' Last Exam
  • Sol: lower-latency everyday tier at reduced price point
  • Astra input at $8/M tokens, $40/M output — 20% cut from prior release
Why it mattersTwo simultaneous tiers at lower prices marks a shift from benchmark maximalism to economic competition with Anthropic and Google.

Anthropic IPO prices above $1T valuation, begins trading as ANT

Anthropic closed its IPO order books nearly seven times oversubscribed, with final pricing targeting a $1 trillion-plus valuation — the largest tech IPO in history. The offering pairs a public tranche with $26B in private placements anchored by Google and Nvidia at $10B each.

  • Ticker ANT on Nasdaq; seven-times oversubscribed at upper $80 range per share
  • Private placements: Google $10B, Nvidia $10B in Class N non-voting shares
  • Average Anthropic employee equity: $72M at IPO midpoint per Bloomberg
Why it mattersThe IPO locks in Nvidia and Google as the financial infrastructure behind Anthropic's compute bets — and prices AI frontier work at a scale that makes the next wave of commitments easier to justify.

Nvidia exits China H200 market; board rules out Huang return

Interim CEO Colette Kress confirmed Nvidia will no longer sell H200 chips in China, ending the tariff-based deal previously brokered by Huang and Trump. Board chair Harvey Jones categorically rejected reinstatement speculation. Nvidia also announced multi-year Vera Rubin supply agreements with TSMC, Samsung, Amkor, and Foxconn.

  • H200 China exit ends the 25%-tariff arrangement negotiated by Huang
  • Board chair: accepted voluntary resignation; moving forward — Huang reinstatement closed
  • Vera Rubin NVL72 supply chain locked with TSMC, Samsung Electro-Mechanics, Foxconn
Why it mattersChina exit removes Nvidia's last significant hardware revenue path there; Vera Rubin supply deals confirm the hardware roadmap survives the leadership disruption.

Cerebras launches disaggregated inference for sub-millisecond latency

Cerebras announced a disaggregated inference architecture that separates prefill and decode compute across its wafer-scale cluster, targeting real-time agentic workloads that require response latencies below one millisecond.

  • Separates prefill and decode stages across distinct wafer-scale compute pools
  • Targets real-time agent applications where token latency beats throughput
  • Available via Cerebras Cloud API starting today
Why it mattersSub-millisecond inference opens a category of live, interactive agent applications that GPU inference cannot economically serve at the same latency.

Microsoft Ignite: Copilot declared the new OS for work, Office apps go GA inside it

Microsoft CEO Satya Nadella declared Copilot the new OS for work at Ignite, as Word, Excel, and PowerPoint creation capabilities became generally available inside the assistant — moving document creation out of standalone apps and into the AI interface.

  • Full Office creation (Word, Excel, PowerPoint) now GA inside Copilot Work tier
  • New OS for work framing positions Copilot against emerging agent OS competitors
  • Same release extends GA to Teams, SharePoint, and Power BI integration
Why it mattersGA status turns Copilot from an add-on into the primary interface for the world's largest enterprise software stack — a structural shift in how a billion Office users will work.

One-third of Gen Z relies on AI for companionship, Gallup finds

A Gallup survey of 15,925 people aged 15-22 across 16 countries found 33% rely on AI for companionship — rising to 50% among daily users — the first large-scale cross-national data on AI as a social substitute.

  • 33% overall; 50% among daily AI users across 16 countries surveyed
  • 15,925 respondents aged 15-22 — largest cross-national dataset on AI social use
  • Survey found no correlation with reduced real-world social relationships in the data
Why it mattersThe numbers scale a question that was previously only anecdotal: AI companionship is already mainstream for the demographic that will define consumer AI for the next decade.
Be subscriber #028 weekly AI email interest list
Past days

The Daily Archive