Dark low-key still-life of a matte black gavel resting beside a blank sealed statute folio with a single cyan ribbon accent. No people, no logos.

The Wednesday AI brief: Sanders moves to ban ASI, Anthropic’s cheaper Opus meets OpenAI’s half-price Sol, and AWS clears NATO Restricted

September 23, 2026 Aaron Harme

In this Brief

Anthropic shipped Opus 5.5 after Amodei’s pace-the-frontier call, and OpenAI halved Sol and Luna the same afternoon
Sanders and Casar unveil a bill to ban ASI, pause advanced AI, and create a Department of Artificial Intelligence
AWS is the first cloud cleared to handle NATO Restricted data in every alliance member state
SoftBank’s junk-bond book for the OpenAI bet draws more than $20 billion of preliminary demand as yields rise
Cisco Talos open-sourced CAIRN and flagged CLOSEDQUORUM malware that polls four models with no human in the loop

Anthropic’s cheaper Opus and OpenAI’s half-price Sol turned the slowdown week into a price war

Setup: Anthropic launched Claude Opus 5.5 on Tuesday, the first model it has released since CEO Dario Amodei said the industry should “pace the frontier,” his call to slow how fast the top labs push the most capable models, The Verge reports (Emma Roth). The company says Opus 5.5 costs 40 percent less to run than Opus 5 while matching Fable 5.1 performance “on most work,” with cybersecurity prompts routed to Opus 4.8 and flagged biology prompts to Opus 5.

Anthropic’s blog puts token prices at $4 / $20 per million input and output versus Opus 5’s $5 / $25, with cache reads at $0.20 per million and output more than 30 percent faster. Anthropic also claims 85 percent fewer boundary circumvention attempts versus Opus 5 or Mythos 5.1 during testing, a company claim.

About 90 minutes later, The Next Web and TechCrunch note OpenAI cut GPT-6 Sol and GPT-6 Luna API prices in half: Sol to $2 / $10 from $4 / $20, Luna to $0.10 / $0.50 from $0.20 / $1.20. OpenAI says Sol makes about half as many mistakes on an internal flagged-chat eval, a company claim.

The essay asked labs to slow the climb. Tuesday’s calendar answered with a cheaper Opus that hands the sharpest prompts to weaker siblings, and an OpenAI mid-stack cut before dinner. That is the slowdown that showed up as a price war.

Longer cut: Anthropic ships Opus 5.5 ten days after Amodei asked labs to pace the frontier.

Sanders and Casar move to ban ASI and stand up a Department of AI

Setup: Sen. Bernie Sanders (I-Vt.) and Rep. Greg Casar (D-Texas), chair of the Congressional Progressive Caucus, are unveiling Wednesday legislation to ban artificial superintelligence, pause advanced AI until guidelines exist, and create a Department of Artificial Intelligence, ABC News reports, carrying AP’s Joey Cappelletti (Sep. 23, 5:03 a.m. ET). AP got the bill first. Multiple employees at leading AI companies are endorsing.

AP defines ASI as AI that exceeds human cognitive ability, or that has enough capability to plan and execute the destruction or disempowerment of humanity. The bill would temporarily pause most advanced systems until new federal safety rules exist, then require federal approval before deploy, with violations up to 20 years in prison in some cases.

Sanders said, “It doesn’t take a genius to say, ‘slow it down.’” AP notes Trump has resisted oversight and is “mainly concerned about the economic implications of a slowdown,” and restates Trump’s Tuesday UNGA push to rename AI as “super intelligence” in U.S. documents. Endorsers via AP include Juan Felipe Cerón Uribe of OpenAI Safety Systems and Swante Scholz of Google DeepMind (speaking for himself), with long odds in a Republican-controlled Congress.

A ban, a pause, and a cabinet-style agency for safety rules before advanced systems ship. Distinct from Tuesday’s rename fight, and written for the systems that could outthink and disempower people.

AWS is first cloud cleared for NATO Restricted across the alliance

Setup: AWS is the first cloud provider approved to handle NATO Restricted (NR) data in all alliance member states, The Next Web reports. The path ran through AWS compliance with NATO directive D32, an assessment by Spain’s National Cryptologic Center (CCN), and NATO approval of those findings for every ally.

NR is sensitive alliance information below top secret. NR-capable systems can run in any AWS region inside NATO countries, and AWS has 15 such regions, seven on mainland Europe. Spain also cleared AWS Europe (Spain) for Difusión Limitada (DL), Spain’s NR equivalent, with data centers in Aragón.

The clearance covers AWS infrastructure. Customer organizations must still accredit their own systems, and each government keeps its national process, so alliance-wide approval is a shared baseline while each capital still runs its own buyer paperwork. Dylan Browne of the NATO Communications and Information Agency told TNW the alliance can use commercial tech securely when the checklist clears, with prior Spanish steps including ENS High and 28 services in the CCN catalog.

Commercial cloud passed the alliance security checklist. National paperwork for each buyer still sits with each capital.

SoftBank’s OpenAI junk-bond book draws >$20B demand as yields climb

Setup: SoftBank saw more than $20 billion in preliminary demand for a junk-bond sale, high-yield debt from a below-investment-grade borrower, and one of the biggest junk-bond deals ever, Japan Times reports, citing Bloomberg (Sep. 23). Pricing is expected Thursday. The interest is early and nonbinding.

SoftBank is seeking more than $11 billion, including $10 billion in dollars and €1 billion (about $1.1 billion) in euros, people familiar told reporters Monday, and SoftBank has committed close to $65 billion to OpenAI. Japan Times on Sep. 21 put dollar proceeds across three tenors and euro proceeds across two maturities, with part aimed at a follow-on OpenAI investment expected next month. A SEDaily paraphrase of Bloomberg and Reuters says proceeds go first to SoftBank’s $10 billion third-tranche OpenAI payment due by Oct. 1.

SoftBank has issued $15 billion of bonds year-to-date in 2026, the largest speculative-grade borrower so far this year. SoftBank 2031 dollar bonds yield 8.2% this month, from as low as 6.7% in January, while credit-default swaps sit at a three-year high and S&P rates SoftBank BB+. Citi and JPMorgan are bookrunners, with pricing aimed at the 24th and settlement around the 29th.

Rising yield means investors want more interest to fund the AI bet. Book-building is underway. Coupons and final size are still unset.

Cisco Talos open-sourced CAIRN and found malware that polls four models alone

Setup: Cisco Talos open-sourced CAIRN, a tracker for AI-enabled malware, and highlighted CLOSEDQUORUM, Windows malware that polls DeepSeek, Qwen, Mistral, and Gemini with no human input to steal logins and crypto, WIRED reports. Lead researcher Ryan Fetterman is named on the project.

Talos says it has catalogued roughly 20 additional AI-malware examples. Developer and real-world use remain unconfirmed.

Four models on a polling loop, no operator in the chat. An open tracker from a major vendor, and a sample that treats frontier APIs as unpaid co-conspirators.

Longer cut: Cisco Talos open-sourced an AI malware tracker and found one that polls four models.

Also noted

Trump told the UN General Assembly on Tuesday the U.S. will rename AI to “super intelligence” in official documents, The Verge reports. Distinct from today’s Sanders ASI ban bill.
Microsoft led a disruption of EvilTokens, an AI inbox-fraud SaaS that hit about 12,000 Microsoft accounts across roughly 10,000 organizations, with $1,500 upfront plus $500 a month and two UK Met arrests on suspicion, Ars Technica reports.
ShinyHunters claim a PeopleSoft zero-day path into AWS GovCloud holding agent and applicant PII, with a roughly 5,000-person sample, a jobs-site defacement, and an FBI statement that it is investigating FBIjobs.gov claims, 404 Media reports.
Peloton set an Oct. 1 launch for three AI treadmills, including a $2,195 fold-up Flex and Vision models with camera Run Analysis for form coaching, Engadget reports.