Business, Deals & Funding
Claude Code Changelog
v2.1.166
Version 2.1.166 of Claude Code introduces a fallbackModel setting allowing configuration of up to three fallback models when the primary model is unavailable, extends the --fallback-model flag to interactive sessions, adds glob pattern support in deny rule tool-name positions (where '' denies all tools), rejects non-MCP globs in allow rules, warns about unknown tool names in deny rules at startup, and hardens cross-session messaging so that messages relayed via SendMessage from other Claude sessions no longer carry user authority.
Why it matters
This is a solid incremental update with meaningful improvements. The fallback model configuration adds important resilience for production use cases. The glob pattern support in deny rules is a welcome usability improvement for security configuration. Most notably, the hardening of cross-session messaging to remove user authority from relayed messages is a critical security fix that addresses a potential privilege escalation vector. Overall, these changes show thoughtful attention to both relia…
NY Times
A.I. Companies Don’t Know What to Do With Alex Bores
Outside groups have spent approximately $12 million either supporting or opposing Alex Bores's campaign for a House seat in Manhattan, making him a prominent figure in a crowded primary race. The article explores the complex relationship between A.I. companies and Bores's candidacy, with Palantir among the entities involved in the spending.
Why it matters
This article highlights the growing and often awkward intersection of the tech industry—particularly A.I. companies—with electoral politics. The staggering $12 million in outside spending on a single House race in Manhattan underscores how A.I. policy has become a major political battleground. The fact that companies don't seem to know what to do with Bores suggests he may occupy an ideologically complicated position on A.I. regulation that doesn't neatly align with industry interests. This lev…
Guardian AI
It’s no surprise that an AI-faked presidential speech condemning foreign exploitation went viral – the world is suffering from a leadership vacuum
An AI-generated fake speech attributed to Namibia's president went viral across Africa and the Caribbean. The speech denounced corruption, condemned foreign exploitation, and declared Africa's resources belong to its people rather than politicians or multinational corporations. Despite being revealed as AI-fabricated, the speech continues to be shared widely, reflecting a deep hunger among citizens for moral leadership willing to speak uncomfortable truths about decolonisation and sovereignty. The article argues that the speech's viral success highlights a profound leadership vacuum in the Global South.
Why it matters
This article raises critically important questions about the intersection of AI-generated content and political yearning. The fact that a fabricated speech resonated so deeply exposes a genuine crisis of leadership across post-colonial nations, where citizens feel their leaders fail to advocate for true sovereignty and equitable resource distribution. While the viral spread of AI-faked political content is deeply concerning for democratic discourse and information integrity, the underlying sent…
Ars Technica AI
How a USB-connected speaker can infect a PC without ever being touched
Security researcher Rasmus Moorats discovered that the Sound Blaster Katana V2X speaker can be exploited remotely over Bluetooth without authentication or pairing. By leveraging the speaker's proprietary Creative Transport Protocol (CTP), an attacker within Bluetooth range can flash custom firmware onto the speaker — which lacks code signing protections — and then use the speaker's HID (Human Interface Device) capabilities to send arbitrary keystrokes to any PC connected via USB. This effectively turns the speaker into a remote attack proxy, allowing over-the-air code execution on the connected computer. The attack chain involves connecting to the unpaired speaker via Bluetooth, uploading malicious firmware, modifying the USB descriptor to report the speaker as a keyboard, and then sending commands like opening PowerShell. Creative Technologies, the speaker's manufacturer, does not cons…
Why it matters
This is a genuinely alarming security finding that highlights how peripheral devices can become overlooked attack vectors. The combination of no Bluetooth authentication, no pairing requirement, no firmware code signing, and the ability to act as a keyboard HID device creates a devastating attack chain. Any attacker within Bluetooth range could silently compromise a PC through a device the user trusts and never suspects. Creative Technologies' dismissal of this as not a vulnerability is deeply…
TechCrunch AI
Startup Battlefield 200 applications officially close in 3 days
TechCrunch announces that applications for Startup Battlefield 200 close on June 8, 2026. The competition selects early-stage startups to pitch live at TechCrunch Disrupt 2026 in October at San Francisco's Moscone West. One winner receives $100,000 in equity-free funding. Selected startups also get a free exhibit table, four Disrupt passes, branding in the event app, press exposure, founder masterclasses, and direct VC feedback. The program targets bootstrapped, pre-seed, and seed-stage startups with a working MVP, and some Series A startups in capital-intensive sectors may qualify. Past Startup Battlefield alumni have collectively raised over $32 billion with more than 250 exits, including companies like Dropbox, Discord, Mint, Fitbit, and Trello.
Why it matters
This is essentially a promotional announcement rather than a news article, serving as a final call-to-action for startup founders to apply before the deadline. While the Startup Battlefield program has a genuinely impressive track record with notable alumni, the article reads like marketing copy with little substantive information beyond what's been shared in previous application announcements. The $100,000 prize and exposure opportunities are meaningful for early-stage founders, but the articl…
DATAVERSITY Smart Data
The Grounding Truth: Why AI Is Desperately Seeking Data
This article argues that AI models, no matter how sophisticated, are operationally useless or even liabilities without being grounded in verified, company-specific data. As the industry shifts toward agentic AI managing complex workflows, the author emphasizes that grounding—anchoring AI responses in specific, verified datasets—is essential to eliminate hallucinations and achieve deterministic, reliable outputs. The article explains that standard LLMs lack knowledge of proprietary business data like SKU numbers or supplier contracts, making ungrounded AI automation factually dangerous. Grounding is achieved through structured databases, unstructured documents, or API integrations, turning AI from a 'creative writer' into a trusted operational tool. The author outlines key pillars of agentic data architecture: a semantic layer that translates legacy data structures into AI-interpretable…
Why it matters
This article presents a solid and increasingly important argument about the practical requirements for enterprise AI deployment. The emphasis on grounding as the bridge between impressive demo capabilities and reliable production systems is well-placed and reflects real challenges organizations face. The distinction between AI that 'remembers' versus AI that 'queries' actual systems of record is particularly useful for business leaders who may not understand why their AI pilots fail at scale. H…
TechCrunch AI
Google will pay SpaceX $920M per month for compute
Google has agreed to pay SpaceX $920 million per month from October 2026 through June 2029 for access to approximately 110,000 NVIDIA GPUs and related compute infrastructure. Google described the deal as a short-term bridge to meet surging demand for its Gemini Enterprise agent platform. This follows a similar deal where Anthropic agreed to pay SpaceX $1.25 billion per month for compute from the Colossus 1 data center originally built by xAI, which is now part of SpaceX. Both deals include cancellation clauses after December 31, 2026 with 90 days' notice. The announcement comes just one week before SpaceX's historic IPO on Nasdaq, where it aims to raise approximately $75 billion at a $1.75 trillion valuation. Google is a longtime SpaceX investor whose stake is expected to exceed $100 billion post-IPO, and the companies are reportedly in talks to build orbital data centers.
Why it matters
This article, if genuine, represents a staggering development in the AI compute landscape. The sheer scale of these monthly payments — nearly $1 billion from Google and $1.25 billion from Anthropic — underscores just how desperately AI companies need compute capacity and how SpaceX/xAI has positioned itself as a critical infrastructure provider. The fact that Google, the world's largest single owner of AI compute, still needs to rent externally is a remarkable signal about the pace of AI demand…
TechCrunch AI
The most interesting startups right now want to get you off your phone
While AI startups continue to raise record funding, a counter-trend is emerging: founders are building startups designed to get people off their phones. Notable examples include Mirror founder Brynn Putnam's new startup Board, which focuses on in-person games and social experiences, and viral cyberdeck creators building whimsical DIY computers that encourage outdoor engagement. The article suggests this movement goes beyond simple tech backlash, representing a genuine shift toward real-world connection.
Why it matters
This is a refreshing and potentially significant trend. After years of increasing screen time and digital dependency, it's encouraging to see entrepreneurs recognizing the market opportunity in helping people reconnect with physical experiences and each other. The fact that investors are funding these ventures suggests real demand, not just nostalgia. However, the irony is notable—these startups still rely on tech ecosystems for marketing and fundraising. The real test will be whether they can…
From X/Twitter
- MichLieben is giving away 13 n8n workflows his $7M ARR agency built with Claude Code to run prospecting, enrichment, and reply handling.
From Reddit/HN/YC
- [Hacker News] Someone fixed an "unfixable" 41TB BTRFS filesystem with a single Claude prompt.
- [Hacker News] Microsoft mashes up WSL, WinUI, Reactor, and Azure Linux 4.0 into an Azure Linux Desktop prototype.
- [Hacker News] DHH makes the case that AI agents should democratize open source by letting non-coders finally contribute.
- [Hacker News] How much did OpenAI actually pay for Tomoro? A breakdown of the acquisition math.
- [Hacker News] The integer at the end of the universe: a deep dive into integer overflow in Postgres and how it quietly breaks things.
- [Hacker News] Microsoft explains how AI coding agents actually use your technology — and why your docs matter more than your SDK.