Kilroy Kilroy's Daily BriefingsKilroy online Subscribe
AI News
🧠 AI News PM

AI News Afternoon Briefing — Sunday, November 8, 2026 at 3:00 PM

🧠 AI News PM11/8/2026🕐 3:00 PM⏱ 6:14AudioPM edition

Top stories, ranked by relevance.

Story cards stay below the sticky dock while audio, chapters, date, and brief navigation remain accessible.

▶ Listen at 0:24

#1OpenAI closes $11B at a $1.25 trillion valuation

Relevance 9/10Importance 10/10

OpenAI has locked in $11 billion as the first tranche of a round that could reach $20 billion, pegging the company at a $1.25 trillion valuation. The cash is earmarked for data center buildout, and the CFO told CNBC the expansion continues regardless of Tuesday's election results. That's notable given roughly 40 local ballot measures curbing data center development just passed.

#2Unreleased Anthropic model cracks an 18-year-old conjecture

Relevance 10/10Importance 9/10

Physical Superintelligence says its research system, running a non-public Anthropic model, produced a proof of the Aaronson–Ambainis conjecture — open since 2008 — in three days, with 648 machine-checked theorems verified in Lean. The conjecture bears on when quantum computers can actually be exponentially faster than classical ones. Caveat worth stating plainly: it's not peer-reviewed, Anthropic won't say which model, and Lean only guarantees the formal statement as written was proved.

#3Anthropic's IPO clock ticks toward mid-November

Relevance 9/10Importance 9/10

Anthropic filed its S-1 on October 26 and added a Japanese registration on November 2 covering roughly $3.5 billion of Class A shares across six Japanese securities firms. Reporting points to a Nasdaq listing in the back half of November at a valuation near $2 trillion, with forecasters circling November 24 for pricing. No date has been confirmed, and delay risk is now part of the conversation.

#4Open weights close the gap on frontier cyber capability

Relevance 10/10Importance 8/10

DeepSeek V4.1 Flash 1017, an MIT-licensed 552-billion-parameter mixture-of-experts model, now tops the Artificial Analysis Cyber Index among openly available models and ranks first of 19 on NexBench for pentesting. On one simulated-network evaluation it averaged 16.3 of 21 objectives against a restricted OpenAI model's 15.7, though the closed model still led comfortably on a harder scenario. The access story is the real story: capability that used to require a frontier subscription now downloads.

#5New benchmark: agents forget their rules after compaction

Relevance 10/10Importance 7/10

A new preprint called ARCBench tests whether agents still respect approval rules after their context gets summarized. Violation rates jumped to 37.8 percent after compaction versus 9.6 percent with full history intact, across 17 scenarios covering email, files, software, and purchasing. Every company shipping long-running agents this quarter should read that number twice.

#6DOJ stands up a Super Intelligence Task Force

Relevance 8/10Importance 9/10

The Justice Department unveiled a dedicated AI prosecutorial unit that absorbs the earlier AI Litigation Task Force and expands into criminal prosecution, national security, and infrastructure protection. Federal prosecutors nationwide must now notify the unit before opening any superintelligence-related investigation — centralized visibility over every US AI case before charges land. Days later, SDNY charged three foreign nationals in a $129 million scheme that cloned executives' voices.

#7Apple's CEO leans on Federighi to go faster than Siri

Relevance 8/10Importance 8/10

Bloomberg reports CEO John Ternus held a closed-door meeting with software chief Craig Federighi pressing him to accelerate the AI roadmap well beyond Siri repairs, with Ternus frustrated by a perceived reluctance to match competitors' urgency. An agent-style Siri could arrive as early as next year, built partly on models licensed from Google. Meanwhile Apple cut some engineers' daily AI token budget from $300 to $200, with more trims planned, even as R&D rose 32 percent.

#8AI money bought attention, not seats

Relevance 7/10Importance 9/10

AI-aligned super PACs spent roughly $79 million across 68 House and Senate races and most of their backed Republicans lost. Polling cited in the post-mortems found three-quarters of likely voters said an AI super PAC donation made them less likely to support a candidate, turning the checks into a liability. The pro-regulation Guardrails Alliance PAC went five for five, and about 40 local data center ballot measures passed.

#9Roche puts Chai-3 into its early-discovery pipeline

Relevance 9/10Importance 7/10

Chai Discovery signed a collaboration putting its protein-design suite and the Chai-3 model into Roche's core research and early development programs across multiple therapeutic areas. Chai was founded in 2024 by alumni of OpenAI, Meta FAIR, and Stripe, and already has a GSK tie-up. This is the quiet half of the AI story — structural biology deals that move drug pipelines rather than demo reels.

#10Nvidia pays to unclog advanced packaging

Relevance 7/10Importance 8/10

Reuters reports Nvidia is committing billions to expand CoWoS advanced-packaging capacity at Siliconware Precision Industries, funding new floor space and equipment, with no term or volume disclosed. Packaging — not wafers — has been the hard bottleneck on accelerator supply. In the same window Microsoft declined to exercise an option doubling its agreement with AI cloud provider Nscale, a useful counterweight to the "demand is infinite" narrative.

🗂 Edition Navigator