Daily AI Briefing — 2026-11-07

Good morning. Here's your AI intelligence for Saturday, November 07, 2026.

The workforce reckoning

New Goldman Sachs research puts hard numbers on what happens to workers displaced by AI. The answer is worse than most economists predicted: workers pushed out by AI automation are 10 percentage points more likely to still be unemployed a year later compared to peers laid off for other reasons. And those who do find new work are largely landing in roles with equal or greater AI exposure — running directly into the same risk again. The data is a warning that retraining rhetoric has outpaced retraining reality.

Rockwell Automation is building its largest factory ever in Wisconsin, outfitting it with AI, robotics, and automation from day one. The move works on two levels: it adds domestic manufacturing capacity and functions as a live demonstration of Rockwell's own industrial AI stack. For a company selling automation solutions to manufacturers, running its own operations on that platform is a credibility argument no marketing budget can replicate.

Medicine and consent

Microsoft's GPT-5-based medical AI achieved nearly double the diagnostic accuracy of emergency room physicians on structured clinical cases — a margin wide enough to accelerate real deployment conversations. The catch is consent: 91% of patients in a parallel survey say AI should require their explicit approval before participating in a diagnosis. That's near-consensus, and healthcare systems will have to build around it. Speed of AI capability and speed of patient trust are on different timelines.

Encryption under pressure

Google DeepMind researchers identified two significant weaknesses in RSA-1024 cryptosystems this week: one that makes signature forgery 1,600 times more efficient, and another that cuts the cost of batch key factoring by 30 times. RSA-1024 is already deprecated by security standards bodies, but it remains in production use across legacy systems worldwide. These findings don't break it today — but they close the gap considerably and compress the migration timeline for organizations still running it.

The UK launched the Policing AI Threat Hub (PATH), the G7's first dedicated AI criminal response unit, prioritizing deepfake detection as AI-generated fraud and synthetic identity crime surge across British law enforcement caseloads. Standing up a permanent unit rather than a temporary task force signals that UK law enforcement views AI-enabled crime as a structural feature of the threat landscape, not a passing wave.

Law and practice

The New York State Bar Association adopted formal AI ethics guidelines — the most comprehensive state bar rules on the subject yet. Attorneys must now disclose when AI tools produced billing efficiencies, protect client confidentiality when using AI platforms, and comply with court-specific AI disclosure requirements. Law has been slow to formalize AI governance. New York's framework will likely become the reference point for state bars across the country working through the same questions.

Physical AI expands

Applied Intuition — whose simulation and validation software is embedded in 18 of the top 20 global automakers — is extending its physical AI platform into heavy freight trucking and defense ground vehicles. The company's core argument is that automotive-grade safety validation infrastructure is the right foundation for any autonomous vehicle, regardless of sector. The defense angle is worth watching: rather than building bespoke simulation systems, defense programs are increasingly pulling from automotive AI development stacks that have already been tested at civilian scale.

Models

Thinking Machines Lab released Inkling 2 Small, a 594-billion-parameter mixture-of-experts model optimized for coding, vision, and autonomous agentic tasks. The inference-efficient MoE architecture activates only a fraction of its parameters per token — making frontier-scale capability accessible at lower compute cost. It enters an increasingly crowded open-weight tier where the competitive question is no longer whether a model is capable, but whether its efficiency profile fits the economics of real deployment.

Schools and workload

New Bett research on UK teachers reveals a persistent gap between AI adoption and actual impact. Teacher use of AI tools has risen sharply in 2026, but the majority report no meaningful reduction in working hours. This mirrors a pattern emerging across professional sectors: AI tools get adopted for specific tasks while the broader administrative load restructures around them rather than shrinking. The question isn't whether teachers are using AI — they are. The question is whether the tools are being aimed at the right friction.

Finance and infrastructure

Nine fintech companies raised $454.5 million across 12 deals in a single week, with AI-focused startups leading every category. The velocity of institutional capital into AI fintech is accelerating even as regulators tighten oversight in financial services — a signal that investors are pricing in compliance burden and betting that well-structured AI applications will clear the bar.

The Novo Nordisk Foundation committed up to €13.3 million to AI4GRIDS, a Nordic cross-border research initiative targeting power grid computation with a goal of running grid simulations 100 times faster than conventional solvers. As renewable energy sources introduce more variability into European grids, real-time AI-assisted planning becomes a stability requirement rather than a research ambition. Nordic power systems — already running high percentages of hydro and wind — are a natural laboratory.

ZTE and Ethio Telecom activated Africa's first pre-commercial AI-powered 5G Radio Access Network. AI-native RAN enables dynamic resource allocation that reduces infrastructure costs at the edge — which matters most in emerging markets where coverage economics have historically constrained rollout. It is a proof-of-concept deployment, but the first deployment is always the one that gets cited in every contract negotiation that follows.

What to watch today

The gap between AI-displaced workers and AI-created roles is becoming a durable political pressure point. Watch for more policy responses to Goldman Sachs-style displacement data as economic anxiety around automation finds its way into legislative agendas.

The consent question in medical AI is unresolved. Health systems, regulators, and AI developers will need to define what explicit patient consent looks like in clinical workflows — it won't stay a survey finding for much longer.

RSA-1024 migration velocity is the security story to track. Teams with legacy dependencies should treat the DeepMind findings as a forcing function for timelines they have been stretching.