Thursday, October 15, 2026

AI credit stress hits records while adoption goes universal and SMB agent money flows

Credit default swap spreads on Oracle, SoftBank and CoreWeave hit record levels as investors question whether debt-funded data centres can generate returns fast enough, with CoreWeave losing a fifth of its value in a day. The same day the FSB finalised 12 AI governance practices for banks and Wharton reported near-universal AI use among large US companies. On the small-business side, Dapta raised $6M for an SMB sales agent and Meta published research showing most owners find AI makes them more productive but want more guidance.

Sources read
20
Claims that stand
8
Quotes found in the source
15 of 15
Claims struck
0
Corrected or withdrawn
0

We quote it or we drop it. Every claim below carries the source’s own words, checked against the text we read.

Since 14 Oct

What we told you last time, ruled on from today’s evidence.

  • Gone quiet
    Nothing new today on these. A limit of what we read, not a verdict.

    JPMorgan blocks outside AI agents from operating customer accounts · Nvidia takes stakes in four French startups including Doctolib · Hugging Face breach becomes the reference case for agent containment · Tabby launches AI accountant as only 11% of SMBs trust AI payroll · Apple launches $299 HomeView, betting on Google-assisted Siri · Pluto Security raises $42M to police employee AI tool sprawl · Taiwan draws US defense startups and chip-veteran deep tech founders · AI guardrails hit the Wisconsin governor race as Cerebras CEO attacks doomers

Every correction we have made →

Each mark on a claim’s line is an outlet, placed by when it published: 16:15 to 06:15 UTC.

AI

  1. AI default insurance costs hit records as CoreWeave debt tumbles

    Five-year CDS spreads on Oracle widened to a record 288 bps, SoftBank to 424 bps and CoreWeave past 1000 bps, with CoreWeave shares down 15% on Wednesday and 2032 bonds trading at 77 cents. The AI buildout is debt-financed, and this is the clearest market signal yet that lenders doubt data-centre revenue will arrive fast enough to service the interest.

    The cost of insuring against US tech defaults has hit record levels in the latest sign of credit stress building in the AI trade

    1 more receipt

    CoreWeave, a loss-making data-centre operator which owes nearly $50 billion according to LSEG data, is in the eye of the storm.

  2. FSB finalises 12 sound practices for AI adoption at financial institutions

    The Financial Stability Board published its final report on responsible AI adoption, covering board-level governance, lifecycle risk assessment, human oversight, and third-party AI dependencies. This becomes the reference framework supervisors will cite when examining how banks deploy models and agents, and it explicitly names third-party AI vendors as a managed risk.

    The Financial Stability Board (FSB) today published its final report on Sound Practices for Responsible Adoption of Artificial Intelligence

    1 more receipt

    The 12 sound practices emphasise the importance of organisation-wide AI governance and highlight considerations for managing risks throughout the AI lifecycle.

  3. Wharton finds near-universal AI use at large US companies, government lags

    Wharton's fourth annual AI Adoption Report, surveying about 1,400 decision-makers and employees, found 91% use AI more than weekly and 62% daily, with AI fully embedded in core workflows at 28% of large companies. The adoption question is settled in the private sector, shifting the competitive edge to workflow redesign, while public-sector research finds mostly avoidance.

    Researchers at the Wharton School recently found that there is near-universal use of artificial intelligence among large United States-based companies.

    2 more receipts

    The report found that AI is fully embedded in the core workflows of 28% of large companies, and has significantly changed processes for 73% of them.

    Our own two years of research at RethinkAI found mostly avoidance and uncoordinated experimentation.

AI agents that find customers for small businesses and founders

  1. Dapta raises $6M seed for AI sales agent aimed at SMBs

    Dapta closed a $6 million seed round led by Chingona Ventures to expand its AI sales agent for small and medium-sized businesses. Seed capital is still flowing to agents that do outbound sales for small firms, a category where the buyer cannot afford a human SDR.

    Dapta, which builds an AI sales agent for small and medium-sized businesses, has secured $6 million in seed funding led by Chingona Ventures

  2. Meta: three quarters of small business owners say AI boosts productivity

    Meta research across nine APAC markets found most small business owners say AI makes them more productive and helps them reach customers across borders, but most want more guidance. Demand for AI among small firms is real but the guidance gap means the winning products will be the ones that remove setup and judgment, not add it.

    New Meta research across nine APAC markets shows how AI is helping small businesses grow, compete with larger rivals and reach customers across borders.

    2 more receipts

    New economic impact reports from Amazon, TikTok, and Shopify reveal the extraordinary reach of platform-powered entrepreneurs, the growing role of AI

    Digital payments, AI, and automation can make some jobs easier, but they also come

  3. Outlit pitches an AI agent assigned to surveil every B2B customer

    Startup Outlit is deploying agents that continuously monitor B2B companies' customers and compile intelligence dossiers, predicting every company will assign one agent per customer. Customer-finding agents are moving from prospecting to persistent surveillance, which will draw privacy scrutiny and could poison trust in agent-driven outreach.

    The startup Outlit is deploying agents that constantly surveil B2B companies' customers and create detailed intelligence dossiers.

tech startups

  1. Founders pitch for $475,000 at Venture Atlanta as Korean startups court LA

    Hundreds of founders pitched investors at Venture Atlanta for a $475,000 prize pool, while six Korean AI and tech startups presented to US investors at K Founders Night during LA Tech Week. Early-stage capital is still being contested in person and cross-border, even as credit markets sour on the AI infrastructure layer.

    Hundreds of startup founders are pitching their companies face-to-face to big-name investors at Venture Atlanta, all hoping to find someone ready to bet on

    1 more receipt

    Six Korean AI and tech startups showcased their innovations to US investors at the K Founders Night event during LA Tech Week.

  2. IranWire: Iranian security agencies extort money from tech startups

    Documents obtained by IranWire show numerous Iranian businesses, including tech startups, facing confidential and informal financial demands from security agencies. It is a reminder that in some markets the state, not the market, is the largest risk to a startup's cash.

    According to documents and reports received by IranWire, numerous businesses in Iran have faced

Why it matters

The AI trade is now splitting into two markets. At the infrastructure layer, debt-funded data-centre builders are being repriced by credit markets, and CoreWeave's bonds at 77 cents suggest lenders no longer accept future revenue as collateral. At the application layer, Wharton's numbers say adoption is universal, the FSB is codifying how regulated firms must govern it, and seed money is still going to agents that sell for small businesses.

If the credit stress spreads from CoreWeave to Oracle and SoftBank, compute costs could rise or capacity could stall, which hits every agent startup that rents inference. Meanwhile the Outlit pitch shows where customer-finding agents are heading: from outreach to persistent surveillance, a direction that Meta's own research says small business owners are not yet equipped to navigate.

What to watch

  • Whether Oracle and SoftBank CDS spreads keep widening after CoreWeave's 15% drop, and whether any AI data-centre borrower misses a payment or restructures.
  • Whether bank supervisors begin citing the FSB's 12 sound practices in examinations, particularly the third-party AI dependency and human oversight items.
  • Whether privacy regulators or B2B customers push back on Outlit-style per-customer surveillance agents.

Sources read

SourceOutletPublished, UTCQuoted
FSB finalises sound practices for the responsible adoption of artificial intelligence (AI) fsb.org04:152×
Q&A: The “collision” between AI, your brain, and society news.mit.edu04:15read, not quoted
Wharton report finds vast majority of U.S.-based companies use AI thedp.com04:152×
The AI Race Government Is Losing governing.com05:151×
Cost of AI default insurance spikes as data centre confidence wobbles reuters.com05:152×
How to Use AI With English Learners Without Encouraging Cognitive Offloading (Opinion) edweek.org05:15read, not quoted
Exclusive | Bloomberg Doubles Down on Conversational AI wsj.com01:15read, not quoted
AI is the prime suspect in the PS5 jailbreak surge mashable.com03:15read, not quoted
More Than Three Quarters of Small Business Owners Say AI Makes Them More Productive, But Most Need More Guidance about.fb.com01:151×
Scale Without Size: Digital Tools and Platforms Are Expanding the Reach and Power of Small Businesses sbecouncil.org21:151×
"Customer Surveillance" Startup Says "Every Company Will Soon Assign an AI Agent to Every Customer to Watch Their Every Move" futurism.com16:151×
CedarDB wants one database to handle both transactions and analytics, as AI agents raise the stakes techcrunch.com22:15read, not quoted
Dapta Closes $6M Seed Round to Expand AI Sales Agent for Small Businesses theaiinsider.tech18:151×
From tap to pay to AI: How technology is affecting small businesses digitaljournal.com19:151×
Funding Tracker '26: Ours Privacy, Arintra and Happy Health fiercehealthcare.com01:15read, not quoted
Anthropic opens Singapore office, signs AI safety pact with government computerweekly.com06:15read, not quoted
High-stakes pitches: Tech startups compete for $475,000 at Venture Atlanta conference cbsnews.com22:151×
Iranian Security Agencies Extort Money From Tech Startups iranwire.com16:151×
LinkedIn Top Startups 2026: The 50 U.S. Companies on the Rise news.linkedin.com20:15read, not quoted
K Founders Night — Korean AI & Tech Startups Pitch to US Investors in LA en.koreadaily.com05:151×

AI Builder’s Edge

Microsoft's VS Code 1.141 (released Oct. 7) lets developers hand a Codex or Copilot conversation between the CLI, ChatGPT app and the editor with history intact, and adds a grid layout for running agents side by side. Meanwhile OpenAI opened a status-page incident on Oct. 13 for degraded performance in GPT-6.1 Sol Codex, and Anthropic's updated usage policy formalises Claude's ability to exit abusive conversations in Claude Code. Databricks is pushing branching databases as agent-era SDLC infrastructure.

Sources read
6
Claims that stand
4
Quotes found in the source
9 of 9
Claims struck
0

We quote it or we drop it. Every claim below carries the source’s own words, checked against the text we read.

Each mark on a claim’s line is an outlet, placed by when it published: 07:16 to 07:16 UTC.

  1. VS Code 1.141 lets Codex and Copilot sessions move between apps

    VS Code 1.141 lets developers start a conversation in the Codex CLI, ChatGPT app, Copilot CLI or Copilot app and continue it inside the editor's Agents window with full history, with only one app allowed to send messages at a time. The editor is becoming the control plane for multiple vendors' agents, and the practical hack is to fall back to a Copilot model once your ChatGPT usage limit hits rather than stop working.

    the latest update making it easier to move between GitHub Copilot and OpenAI Codex, manage multiple agent sessions and control what those agents can access

    3 more receipts

    Developers can start a chat in the ChatGPT app or Codex CLI, then pick it up in VS Code on the same machine, with the conversation history intact.

    developers can select a Copilot model in VS Code to keep working with their GitHub Copilot subscription after reaching a ChatGPT usage limit

    Only one application can send messages to the conversation at a time.

  2. OpenAI logs degraded performance incident for GPT-6.1 Sol Codex

    OpenAI opened a status-page incident on October 13 covering degraded performance in GPT-6.1 Sol Codex and Work mode. Builders leaning on Codex for long agentic runs should expect slowdowns or failures until the incident clears, which makes the VS Code model-fallback path immediately useful.

    OpenAI opened a status-page incident on October 13, 2026, for degraded performance in GPT-6.1 Sol Codex and Work mode.

  3. Anthropic policy update: Claude can exit abusive sessions, including in Claude Code

    Anthropic's updated usage policy prohibits sustained cruelty toward Claude and notes Claude can already end conversations with abusive users on Claude.ai and Claude Code, alongside tightened rules on weapons, surveillance and elections. Claude Code users who vent at the model in long sessions now have an explicit rule, though Anthropic says ordinary frustration and testing are untouched; the weapons and surveillance clauses matter more for what agent builders can ship.

    Anthropic published an updated usage policy on Thursday that prohibits sustained abusive or cruel behavior toward its Claude AI models

    2 more receipts

    Claude has had the ability to exit conversations with abusive users on Claude.ai and Claude Code since a prior update

    New usage policy cracks down on cruelty to chatbot, alongside slightly more pressing matters of weapons development, surveillance, and election

  4. Databricks pitches Lakebase branching databases for coding agents

    Databricks published a piece positioning Lakebase database branching as infrastructure for an agent-driven software development lifecycle. Branchable databases let each agent run get its own isolated data state, which is the missing piece when multiple agents work in parallel on the same app.

    As coding agents take on a growing share of development work, developers are increasingly shifting toward

Why it matters

The centre of gravity is shifting from which model you use to how you orchestrate several of them. VS Code's session handoff and grid view, plus the Copilot fallback when ChatGPT limits bite, make vendor-switching a routine workflow rather than a reset, and OpenAI's same-week degradation incident shows why that matters.

On the infrastructure side, Databricks framing branching databases as an agent primitive signals that the tooling stack below the editor is being rebuilt for parallel agent runs. Anthropic's policy update is mostly housekeeping for devs, but the explicit weapons-software and surveillance clauses draw lines agent builders should read before shipping.

What to watch

  • Whether OpenAI resolves the GPT-6.1 Sol Codex incident or it widens to other Codex surfaces.
  • Whether the VS Code one-app-at-a-time constraint on shared Codex sessions gets relaxed in the next release.
  • Whether other database vendors follow Databricks with branch-per-agent-run features for coding agents.

Know someone watching this? Pass it on.

20 sources, at least 19 minutes of reading. This briefing took about 3. You’re done.

Counted from the 6 articles we could read in full, at 230 words a minute; 14 more sat behind a paywall or gave us only a headline.