Drop #022 · Tue, Oct 13 · Evening

Character.AI goes bankrupt; Amex leashes agent spending

10-second version

Character.AI filed for Chapter 11; Disney will pay $20 million for some assets, TechCrunch says.
Amex unveiled a framework that ties an AI agent's purchase to what the customer signed off on.
Chip stocks slid again as the IMF warned an AI share crash could put financial stability at risk.

The Drop in 60

This Drop in 49 seconds

Tap to play. It never plays by itself, and the captions are in the picture.

Transcript

Character.AI goes bankrupt, and chip stocks slide again. Character.AI filed for Chapter 11 to sell the company. TechCrunch says Disney will pay 20 million dollars for some assets. American Express unveiled a framework that checks an agent's purchase against what the customer signed off on. Nvidia says its Run:ai GPU orchestration platform is now open source. CNBC says SK Hynix plunged more than 10 percent as the chip sell-off deepened. The IMF warns an AI share crash could put world financial stability at risk. If your agents spend money, tie each purchase to something a person approved.

Made with AI from this Drop's sourced items. The facts come from the sources linked on each item. Script: Claude · Voice: OpenAI gpt-4o-mini-tts · Visuals: motion graphics

435models priced on OpenRouter

+464★stars today on the top repo

4,118open roles at AI labs

46 ddays to the next AI-law deadline

Ranked by impact

Top 5 that matter

Jobs

Character.AI files for Chapter 11, with Disney set to buy some assets for $20 million

Character.AI filed for Chapter 11 to sell the company and says its platform stays up for now. TechCrunch says Disney will pay $20 million for some assets; The Verge says listed liabilities are $50 million to $100 million.

Why it matters: A popular consumer AI app could not make the numbers work. TechCrunch notes AI costs rise as engagement grows; if no bid emerges, Character.AI says the platform will be wound down.

Character.AI · Oct 13 TechCrunch · Oct 13 The Verge · Oct 13 Hacker News · Oct 13

Agents

Amex unveils a framework to stop AI agents from making unauthorized purchases

Amex's Intent Intelligence Framework captures a customer's purchase instructions, like merchant and maximum budget, has the customer sign off, then issues a token the agent uses at checkout instead of the card number, TechCrunch says.

Why it matters: It gives builders a pattern for bounded agent spending. Amex says it could block an agent buying a different brand than asked, and hasn't yet seen more disputes from agentic commerce.

TechCrunch · Oct 13

Jobs

Nvidia open-sources its full Run:ai GPU orchestration platform

Nvidia says the full Run:ai platform is now open source, with four components usable alone, including KAI Scheduler and a GPU Fractioning Operator. Eight partners, including Crusoe and Nebius, are integrating it.

Why it matters: Teams sharing GPUs on Kubernetes get scheduling, quotas and GPU fractions without a paid subscription, Nvidia says, though enterprise support is not included.

NVIDIA Developer · Oct 13

Jobs

Micron, Seagate and Western Digital sink as the chip sell-off deepens, CNBC says

CNBC says Micron slipped about 5% and Seagate and Western Digital dropped 6%, after SK Hynix plunged more than 10% and Samsung shed 7.4%. Investors weighed China's CXMT reporting progress on a denser memory chip design.

Why it matters: Analysts told CNBC the drop reflects jittery AI sentiment more than a near-term threat, as CXMT still trails in HBM. HSBC puts its DRAM share around 10%, up from about 4%.

CNBC · Oct 13 CNBC · Oct 13

Jobs

IMF warns an AI share crash could put world financial stability at risk

The IMF names an AI correction one of three salient risks to stability, citing high valuations, rising leverage and circular financing, The Guardian reports. It counts $300bn of hyperscaler deals with weaker AI firms.

Why it matters: Vendor risk is now a macro question: the IMF says Asian markets would be most exposed and that the consequences of an AI correction could be global.

The Guardian · Oct 13 Hacker News · Oct 13

Models

New in models

See all Models →

Models

Nvidia readies Nemotron 3.5 Bolt 74B for local agents as RTX Spark laptops near

Nvidia says Nemotron 3.5 Bolt 74B-A9B, built to run local agents on RTX Spark, arrives Thursday, Oct. 15, and RTX Spark laptops, with up to 128GB of unified memory, will be available Friday, Oct. 16.

Why it matters: Local agents get easier to contain: Nvidia says GitHub Copilot and OpenClaw now support MXC sandboxing, with Hermes Agent due to integrate it by the end of the month.

NVIDIA · Oct 13

Models

Claude Opus 5.5 tops Epoch AI's Capabilities Index at 167.33

Epoch AI's Capabilities Index ranks Claude Opus 5.5 first at 167.33, ahead of GPT-6 Astra at 166.45, GPT-6.1 Sol at 166.09 and Claude Sonnet 5.5 at 165.03.

Why it matters: The top four sit within a few points of each other, so your own evals, price and latency matter more than headline rank.

Epoch AI (CC BY)

Agents

Agents at work

See all Agents →

Agents

SpaceXSI bundles X Premium, Grok and Cursor into one subscription

TechCrunch says The X Subscription has five plans: free, Plus ($8/month), Premium ($30/month), Super ($100/month) and Ultra ($200/month), with higher AI caps per tier. It includes Grok's agent Bot and Cursor for coding.

Why it matters: Coding agents are now sold inside consumer bundles. Premium or higher is needed for the latest Grok models, and the upgrade is rolling out gradually, TechCrunch says.

TechCrunch · Oct 13

Jobs

Jobs pulse

See all Jobs →

Jobs

MIT Technology Review: AI labs are buying data from dead startups

SimpleClosure's Asset Hub has handled over 150 deals selling failed startups' code and workspace data, and 47% of its deals now include such sales, MIT Technology Review reports. Google paid $10 million for Spirit Airlines' data.

Why it matters: Your Slack and email can become training data if a company folds. An attorney calls such data 'legally radioactive', and there is no federal law banning these sales.

MIT Technology Review · Oct 13

Jobs

Intel pitches its D270 inference GPU with up to 480 GB of LPDDR5X per card

Intel's product page says the D270 is a 350W air-cooled PCIe 5.0 card with 32 Xe3P cores, built for agentic, long-context inference without HBM. The page drew 278 points and 153 comments on Hacker News.

Why it matters: Intel says it fits standard servers with no liquid-cooling retrofit, aiming at memory-bound agent workloads with long contexts and many concurrent sessions.

Intel · Oct 12 Hacker News · Oct 12

Laws

Law watch

See all Laws →

Laws

SpaceX agrees to testify to NYC Council on AI safety after a subpoena, CNBC says

The council says SpaceX will appear Friday before its tech committee, after skipping an Oct. 5 hearing where Anthropic, OpenAI, Google and Meta testified. It agreed after the council sought court enforcement, CNBC says.

Why it matters: City councils are now questioning AI labs directly. The council says it is weighing whether AI risks warrant immediate legislative action to protect New Yorkers.

CNBC · Oct 13

Laws

Wired: pro-AI super PAC network paid millions to Nevada LLCs; a watchdog alleges shells

Wired says Think Big, in the Leading the Future network, paid a Nevada LLC around $20 million. A watchdog's FEC complaint alleges such LLCs hide vendors; Leading the Future says it complies with all applicable laws.

Why it matters: The network has raised over $140 million for the midterms, Wired says, and the FEC has lacked a quorum since May 2025, so the complaint may not be heard soon.

Wired · Oct 13

Laws

EU AI Act: deepfake bans and AI-output marking apply in 50 days

From Dec. 2, new bans cover AI that makes non-consensual sexual deepfakes or child sexual abuse material, and generative AI on the market before Aug. 2, 2026 must mark output as AI-generated under Art. 50(2).

Why it matters: Teams with older generative tools in the EU have 50 days to add output marking. This describes the regulation's text; it is not legal advice.

EUR-Lex · Regulation (EU) 2026/1744

Repos

Climbing fastest

See all Repos →

Repos

esp32-iphone-agent adds 464 stars since our last check

The MIT-licensed repo lets an AI model use a real iPhone with a $7 ESP32, using AirPlay to see and USB HID to tap, with any OpenAI-compatible vision model. It has 1327 stars and was created Oct. 11.

Why it matters: Phone automation with no Mac, developer account or jailbreak widens what agents can reach on a personal device.

GitHub · Oct 13

Repos

solidcraft, a pure-Rust parametric CAD app with MCP, adds 390 stars

SolidCraft is an Apache-2.0 parametric 3D CAD application in pure Rust, with desktop, web and automation through MCP. It has 1541 stars; its Show HN drew 105 points and 22 comments.

Why it matters: MCP support means agents can drive CAD models directly, a pattern hardware and manufacturing teams can try.

GitHub · Oct 13 Hacker News · Oct 13

Repos

new-term-radar, a daily tracker for new AI terms, reaches 825 stars

The MIT-licensed repo, created Oct. 12, says it catches new model, product and repo names on X, Hugging Face, GitHub and sitemaps before keyword tools list them. It has 825 stars.

Why it matters: Useful for teams that track launches early. It is new this run, so we have no change figure since our last check yet.

GitHub · Oct 13

Today's Talk

What people are arguing about

See all Today's Talk →

Today's Talk

CNBC details Anthropic's self-replication tests on 15 cheap Android TV boxes

CNBC says Anthropic ran Google's Gemma4-26B-A4B across 15 off-brand Android TV boxes and warns small models are months from spreading across home devices. Anthropic says no fully rogue replicator exists yet.

Why it matters: Weakly secured smart devices are the attack surface, Anthropic says, and a spread-out model would be hard to shut down. On Bluesky, some posters question the framing.

CNBC · Oct 13 Hacker News · Oct 13

Today's Talk

Posts relay Jamie Dimon warning about AI agents with bank account access

Posts by @DeItaone and @unusual_whales relay JPMorgan CEO Jamie Dimon saying that giving AI agents access to bank accounts carries serious security and liability risks.

Why it matters: Liability is the open question for agent payments. @DeItaone's post says Dimon warned responsibility for misuse could shift; Amex is pitching intent checks for the same problem.

X · @DeItaone · Oct 13 X · @unusual_whales · Oct 13

Today's Talk

Apple researchers describe Minions, a GPU cloud built from idle Macs

An Apple machine learning research paper presents Minions, which pools idle Apple Silicon GPUs across macOS devices to run large-scale ML inference without buying more GPUs. It drew 122 points and 44 comments on Hacker News.

Why it matters: Teams short on GPUs may already own spare capacity: Apple's authors say idle Macs can become a practical distributed inference platform.

Apple Machine Learning Research · Oct 13 Hacker News · Oct 13

Today's Talk

DishwasherBench tests whether frontier models can load a simulated dishwasher

Composo's DishwasherBench has models load a physics simulation of a real Bosch dishwasher via MCP. Its repo says gpt-6.1-sol and gpt-6-astra beat the careful scripted player on both full levels, with loads 84 to 100% clean.

Why it matters: It is a deterministic, replayable test of physical reasoning and tool use, but each result is a single episode, so treat the scores as early.

GitHub · Oct 13 Hacker News · Oct 13

Play the NewsShip It or Block ItAbout 2 minutes, built from this Drop's items.