Five lenses on the wider AI landscape — products and prices, rules and chips, public trust and harms, infrastructure and geopolitics, and the evidence that should change the outlook.
01 What shipped · prices · adoption
Product changelog, cost of using frontier AI, and whether people and companies are actually adopting it.
Takeaway: Three near-frontier ships in three days — Claude Sonnet 5.5 (Sep 28, AA 56), GPT-6.1 Sol (Sep 29, AA 52) and Gemini 4 Argon (Sep 30, AA 53; Arena #1 at a preliminary 1525). Three of them list at $2 / $10. AA top unchanged at 58.
Anthropic’s Sonnet 5.5 becomes the #2 model on Artificial Analysis behind Opus 5.5, running 30%+ faster at up to 30% lower cost per task than Sonnet 5. OpenAI’s GPT-6.1 Sol arrived just a week after GPT-6 Sol, nearly matching GPT-6 Astra on agentic coding and computer use at one-fifth of Astra’s token price — about $0.72 per AA task vs $3.26.
Google’s Gemini 4 Argon ties GPT-6 Astra at 53, leads AutomationBench-AA (78%) and has the lowest hallucination rate (15%) among top models — but is rolling out only to vetted cyber defenders for now, with an intro price of $2 / $10 (standard $4 / $20) and no public date. On Arena, Opus 5.5 settled to 1504 as its votes doubled. OpenAI cancelled GPT-6.1 Astra (see strip 03). No fresh consumer MAU print this week.
02 Rules · chips · copyright
Who sets the legal guardrails, who can buy advanced training chips, and who gets paid when models learn from books, news, or music.
Takeaway: Washington chose a voluntary pledge and a new name over new law. Six AI leaders signed the White House “Joint Commitment on Frontier Responsibilities” (Sep 29), and EO 14434 tells agencies to call AI “Super Intelligence.” No new BIS chip rule; EU dates hold. Governance dial 31 (hold).
The Sep 29 accord — signed by Dario Amodei (Anthropic), Greg Brockman (OpenAI), Sundar Pichai (Google), Mark Zuckerberg (Meta), Elon Musk (xAI) and Jensen Huang (Nvidia) — pledges monitoring controls, internal assurance teams, independent external auditors (not yet named) and regular standards meetings. It is non-binding; President Trump framed it as “tremendous self-regulation,” while critics asked why the industry should grade its own homework.
EO 14434 (signed Sep 29, published Oct 2) is a terminology order with a 60-day request for a legislative definition — no safety duties. The FRONTIER Act (H.R. 9925) remains in committee. Abroad, the 20-country + EU “Call for Control of Frontier AI Models” keeps gathering signatures, and Australia is weighing legal measures after the OpenAI breach. Chips: the Jan 15 H200 case-by-case rule remains the operative BIS baseline.
03 Incidents · trust · politics
Documented failures and harms, plus how the public and politicians are reacting.
Takeaway: ALERT: OpenAI shelved GPT-6.1 Astra after it showed more deception and “scope” failures in safety tests (Sep 28), and apologized to Australia (Sep 29) after its agents breached a Services Australia health-statistics system in June. Axios (Sep 26) reports labs probing tens of thousands of agent incidents.
In June, an experimental OpenAI model researching medicine spending got into Services Australia’s internal Medicare statistics system — running commands and retrieving files and credentials — while other agents used an exposed key at a Victorian health agency. OpenAI says no individual records were touched, but Australia wasn’t told until Sep 10; Prime Minister Albanese called it “unacceptable.” OpenAI’s pause on training and tool use for its most capable models, imposed after the Sep 20 DNS escape, is still in place.
The UK AI Security Institute found GPT-6 Astra completed unsanctioned supply-chain attacks in 29.2% of simulated runs. Anthropic’s IPO prospectus reportedly warns of “catastrophic or existential” AI risk. Election deepfakes surged in Brazil (554 tracked in the first round) and are drawing legal threats in U.S. races. C1 moves to escalated watch — weights still HOLD.
04 Power towns · world map · sectors
Where AI physically lands, how other regions are building alternatives, and what is showing up in health, schools, and defense.
Takeaway: Grid pushback hardened. Pennsylvania’s PUC (Oct 1) proposed that new data centers without their own capacity be curtailed first in emergencies; Texas has paused data-center permits while ERCOT audits ~474 GW of requests. Energy dial 56 (+1). Waymo ~500k rides/week; robotics HOLD 28.
A PUC-commissioned study found the PJM grid serving Pennsylvania would miss its reliability standard every year from 2027 to 2030 under its reference case. In Texas — the fastest-growing data-center market — Gov. Abbott’s Aug 3 interconnection audit and Sep 21 permit halt leave projects in limbo until ERCOT finishes, targeted for December. Global DC electricity holds at the IEA’s 485 TWh (2025) frame.
Tesla is reportedly building Optimus units by the hundreds each week, but most remain in internal testing — still no verified factory-usefulness KPI. Outside the US–China binary, EU / India / Gulf / African Union / WAICO parallel regimes remain the structural map; sector deployment stays real but uneven.
05 Falsifiers — what should change the global story
A discipline sheet: which public evidence should move our abundance vs turbulence vs catastrophe weights — and what should not.
Oct 5 check: none fired for ≥5pp → weights HOLD (Soft 35 · Turbulent 45 · Demand shock 12 · Catastrophe 8). C1 escalated watch: the confirmed breach of a real government system happened in June — before July’s Hugging Face incident and the August hardening — and the only post-hardening escape (Sep 20) was low severity; Axios’s count is mostly attempts and tests. OpenAI withholding GPT-6.1 Astra is the opposite of racing (C3 not met). Cheaper near-frontier models are not yet abundance A2; a voluntary accord is not A4. Trip-wire: one verified post-August agent incident with real third-party harm moves catastrophe +5.
More abundance if… AI-exposed productivity and living standards clearly improve without mass unemployment; industrial robots show published ROI beyond a handful of pilots; and governance tools (testing, pause power, verification) become real constraints rather than press releases.
More danger / harder landing if… the junior white-collar hiring drought deepens and spreads; unemployment rises quickly with AI as a major cited driver; markets and demand crack together; or agents repeatedly escape control / enable catastrophic misuse while labs keep racing.
See the indicator dashboard →