Business, Deals & Funding
Claude Code Changelog
v2.1.174
Version 2.1.174 of Claude Code adds a `wheelScrollAccelerationEnabled` setting for controlling mouse-wheel scroll acceleration in fullscreen mode. It fixes several issues with the `/model` picker: the model family that Default resolves to is no longer hidden (showing Opus or Sonnet appropriately based on plan type), the picker no longer shows a hardcoded Sonnet version label when `ANTHROPIC_DEFAULT_SONNET_MODEL` is set to a different Sonnet, and a misleading 'Fable 5 is now consuming usage credits' banner has been fixed.
Why it matters
This is a minor but quality-focused release. The model picker fixes are important for user clarity—hiding the actual model being used or showing incorrect version labels could cause real confusion, especially for users on different plan tiers. The scroll acceleration setting is a nice quality-of-life addition. The 'Fable 5' banner fix sounds like a quirky bug. Overall, these are sensible polish fixes that improve the user experience without introducing major new functionality.
Claude Code Changelog
v2.1.175
Version 2.1.175 of Claude Code adds the `enforceAvailableModels` managed setting. When enabled, this setting ensures the `availableModels` allowlist also constrains the Default model — if the Default model would resolve to a disallowed model, it falls back to the first allowed model. Additionally, user or project settings can no longer widen a managed `availableModels` list, giving administrators stricter control over which models are available.
Why it matters
This is a meaningful enterprise/admin-focused feature that tightens governance over model usage. It closes a loophole where users could bypass model restrictions through default model resolution or by widening the allowlist in their own settings. For organizations that need strict control over which AI models are used (for cost, compliance, or capability reasons), this is an important addition. It's a small but well-targeted change that reflects maturing enterprise administration capabilities i…
Claude Code Changelog
v2.1.176
This changelog entry for claude-code v2.1.176 describes several updates: session titles now generate in the conversation's language (with a configurable language setting), a new footerLinksRegexes setting was added for regex-matched link badges in the footer, Bedrock credential caching was improved to use the credential's actual expiration time instead of a fixed 1-hour cache, and a fix was made to prevent alias model picks from being redirected to blocked models via the availableModels enforcement.
Why it matters
This is a solid incremental release with practical improvements. The language-aware session titles and improved credential caching are thoughtful quality-of-life enhancements. The availableModels enforcement fix addresses a potential security/policy bypass issue, which is important for enterprise users. The changelog entry appears truncated, which is slightly frustrating, but the visible changes are all meaningful and well-considered.
TechCrunch AI
Andrew Yang thinks the next big startup opportunity is lowering the cost of living
Andrew Yang, entrepreneur and former presidential candidate, believes the next major startup opportunity lies in lowering Americans' cost of living rather than extracting value from them. Inspired by Mark Cuban's Cost Plus Drugs model, Yang launched Noble Mobile, a mobile virtual network operator offering cell service at a fraction of traditional carrier prices and returning unused data savings to customers. Yang argues that as AI displaces jobs and compresses wages, startups that reduce costs for basic needs — housing, food, wireless, transportation, education — represent a rich vein of opportunity. Noble Mobile has grown to thousands of customers with millions in revenue since launching in September 2025, though Yang acknowledges investor reluctance, as capital is heavily concentrated in AI rather than thin-margin consumer businesses with social missions.
Why it matters
Yang's thesis is compelling and timely. The idea that startups can compete by returning margin to customers rather than maximizing extraction is a genuinely interesting counter-narrative to the dominant Silicon Valley playbook. Cost Plus Drugs proved the model can work in pharmaceuticals, and applying it to wireless and other essential spending categories makes intuitive sense, especially as economic anxiety grows. However, the fundamental challenge remains: thin-margin consumer businesses are…
Guardian AI
Dutch far-right party pays damages to court artist after changing image with AI
A Dutch court artist, Petra Urban, received damages after a member of Geert Wilders' far-right PVV party used one of her courtroom sketches without permission and altered it with AI to make the depicted Syrian brothers look more menacing. The manipulated image was used in a party video.
Why it matters
This case highlights the dangerous intersection of AI manipulation, copyright infringement, and political propaganda. Using AI to alter a court artist's work to make defendants appear more threatening is a deeply unethical act that serves to inflame xenophobic sentiment. It's encouraging that the artist received damages, as it sets a precedent for protecting artists' rights and integrity against politically motivated AI manipulation. This incident underscores the need for stronger regulations a…
NY Times
Anthropic Blocks Foreigners From Using Mythos and Fable AI
Anthropic has blocked foreign users from accessing its Mythos and Fable 5 AI systems after the federal government ordered restrictions citing national security concerns.
Why it matters
This report reflects the growing intersection of AI development and national security policy. Government-imposed restrictions on advanced AI systems for foreign users are not surprising given the strategic importance of cutting-edge AI capabilities. While such measures may be necessary to protect national interests, they also raise concerns about fragmenting the global AI ecosystem and could push other nations to accelerate their own AI development independently. The move highlights the tension…
TechCrunch AI
Anthropic’s safety warnings may have just backfired — the government has pulled the plug on its most powerful AI
The U.S. government ordered Anthropic to shut down access to its two most powerful AI models — Claude Fable 5 and Claude Mythos 5 — citing national security concerns related to a claimed jailbreak vulnerability. Mythos 5 had been kept restricted under a program called Project Glasswing due to its exceptional ability to find software security vulnerabilities, while Fable 5 was a guardrailed public version released just three days prior. Anthropic complied but publicly disagreed with the decision, arguing the alleged jailbreak was narrow and non-universal, that similar capabilities exist in competitors' models like GPT-5.5, and that its independent classifier safety systems remain intact. The company warned that applying this standard industry-wide would halt all new frontier model deployments.
Why it matters
This article describes a scenario that, if real, would represent a significant and concerning precedent for AI regulation — a government recall of a commercial AI model based on a narrow jailbreak claim with only verbal evidence provided. Anthropic's frustration seems justified: if the standard for pulling a model is any potential jailbreak, no frontier AI model could survive deployment. However, the government's caution around a model with demonstrated ability to find vulnerabilities in every…
TechCrunch AI
SpaceX IPO: Live updates on everything you need to know
SpaceX has gone public in the largest IPO in history, pricing 555.6 million shares at $135 each to raise $75 billion. Shares opened at $150 on Nasdaq, an 11% pop, and closed at $160.95, up 19% on the first day of trading. The IPO is poised to make Elon Musk the world's first trillionaire. Goldman Sachs and Morgan Stanley led the underwriting, with banks earning about $500 million in total fees. SpaceX COO Gwynne Shotwell mentioned in a CNBC interview that a merger between SpaceX and Tesla 'might make Elon's life a little easier.' Robinhood reported record-breaking trading traffic following the debut. The company, now 24 years old, is known for its reusable rockets and Starlink satellite network.
Why it matters
This is a landmark financial event that validates SpaceX's extraordinary trajectory from a startup with early rocket failures to the most valuable IPO ever. The 19% first-day gain suggests strong investor demand, though the sheer scale of the offering — $75 billion — raises questions about whether the valuation is sustainable or driven by hype and Musk's celebrity. Shotwell's comment about a potential SpaceX-Tesla merger is particularly noteworthy and could signal strategic intentions that woul…
From X/Twitter
- Elvis Sun trained Fable 5 on 1,000+ legendary headlines spanning 100 years to build a headline-generation skill — his argument: you can't teach taste with rules, only examples.
- Extend AI open-sources its CLI so you can parse and extract entire document corpora without leaving the terminal — ships with a built-in agent skill for Claude Code, Codex, and Cursor.
- 9to5Mac breaks down Apple's new Foundation Models: on-device AI, cloud AI, and everything in between.
- Shannon Holmberg argues distribution is the moat now for AI startups — the model and harness aren't the edge anymore, getting in front of people is.
- The $599 Mac Mini M4 + Ollama + Claude Code stack is replacing $4,800/year in subscriptions — one env var redirects Claude Code to your local box.
- Zephyr_hg spent a year reading six morning news sources by hand, then built one Cowork Routine that does the same job before sunrise — 90 minutes a day reclaimed.
From Reddit/HN/YC
- [Hacker News] Lojban, the constructed language designed to be syntactically unambiguous, resurfaces on Hacker News.
- [Hacker News] A case for why you should teach your kids to pirate — not ships, software.
- [Hacker News] LaurieWired makes the case that old games feel "hard" because modern emulators add layers of abstraction the original hardware never had.
- [Hacker News] Drinkner matches your bar to 102 IBA cocktails and only calls AI when it can't find a recipe match.
- [Hacker News] Microsoft's developer blog warns: your agent just scaffolded a project from 2020 and you didn't notice.
- [Hacker News] Addy Osmani on loop engineering — designing the feedback loops that keep coding agents from drifting.