Wednesday, November 4, 2026
Ohio voters banned or restricted AI data centers in at least 14 municipalities as AMD's data center revenue nearly doubled - local resistance and compute demand both intensifying the same day. Goldman Sachs pulled Claude Code and GitHub from traders and bankers, keeping them for engineers only. Separately, Nathan Lambert walked back his own stance on open-weight cyber risk, and Simon Willison ran a 'frontier-exclusive' AI capability on a laptop.
Signals
Weight-space programming isn't frontier-exclusive: a small local model now replicates most of the capability Nov 3's brief called frontier-only. #
Simon Willison ran Rikiya Takehi's weight-reading task on a local 27B Qwen model Nov 3, correctly naming Pac-Man, Pong and Breakout from raw weights, matching Kimi K3.1 Pro and failing only Freeway. He calls the reasoning 'near-incomprehensible' and isn't ready to declare source code obsolete. Simon Willison, Qwen 3.8 27B on my laptop can interpret model weights, too
A leading open-weights voice now concedes he underestimated how much open models could accelerate real-world cyberattacks. #
Nathan Lambert wrote Oct 30 that his April claim - an open-weight Claude Mythos release 'would have been fine' on cyber risk - undersold the disruption since seen in PHP2Shell and the GitLab RCE. He still argues unsecured open-model APIs, not open weights themselves, cause most near-term misuse. Nathan Lambert, Was I wrong about open models & cyber risks?
Culture and process, not model quality, determine whether teams capture AI coding's throughput gains. #
Neil Kakkar's Nov 3 essay, edited by Gergely Orosz, documents a demonstrated case: moving from Bloomberg to AI-native startup Tano took him from 8-20 merged PRs a month to over 200, median merge time under an hour, revert rate 6x lower - from changed review process, not a better model. Neil Kakkar, in The Pragmatic Engineer: How software engineering is changing: a tale of two eras
News
WATCHAMD's data center revenue nearly doubles, guidance jumps to $17 billion #
AMD reported Q3 revenue of $13.4 billion Nov 3, with data center revenue up 97% year over year to $8.5 billion. Q4 guidance rose to about $17 billion, roughly 66% growth. AMD, Third Quarter 2026 Financial Results
WATCHGoldman Sachs pulls Claude Code and GitHub access from traders and bankers #
Goldman cut most non-engineering staff off Anthropic's Claude Code and Microsoft's GitHub tools, keeping access for its roughly 12,000 engineers. A spokesperson cited productivity and risk management, reversing July's embrace of 'vibe coding.' The Decoder, Goldman Sachs pulls AI coding tools from traders and bankers
WATCHApple's FY2026 10-K adds a new risk factor for autonomous AI agents #
Apple's annual filing warns AI agents 'can take actions autonomously,' reaching systems they're not authorized to access and causing harms, like deleted data or completed transactions, that may be impossible to reverse. Apple Inc., Form 10-K FY2026 (SEC filing)
WATCHApple brings early Apple Intelligence features to China via Alibaba's Qwen #
iOS 27.3's first beta, out Nov 4, adds Writing Tools, Genmoji and notification summaries for iPhone 15 Pro and newer in China, powered by Alibaba's Qwen. Siri AI and Image Playground stay excluded. MacRumors, Apple Says iOS 27.3 Beta Coming Next Week
The long view
A year ago, opposition to AI data centers worked through town-board moratoriums and quiet delays - Data Center Watch counted $98 billion in projects blocked or delayed in spring 2025, but no US voter had directly decided one at the ballot. On Nov 3, Ohio ran 21 such measures across 18 municipalities, passing bans or restrictions in at least 14, including Ashville, where residents voted to block a data center campus already under construction. If this holds, developers now face a veto that can land after groundbreaking, not before.
Also noted
- Claude Code deprecates Opus 4.8's fast mode (gone Nov 30) and now confirms before emailing a user's address. Claude Code releases #
- Anthropic's SDKs shipped GA spend-limit and RBAC controls Nov 2, letting orgs cap AI spend per user. anthropic-sdk-python v2.4.0 #
- Zvi: OpenAI's revised Model Spec moves harm-prevention authority to the user level and drops its 'love humanity' language. Zvi Mowshowitz, OpenAI Updates Its Model Spec #