Tech Traffic: October 5
The AI Chicks
Tech Traffic
The AI news worth knowing.
Photo: Markus Spiske / Unsplash
Anthropic's leaked IPO prospectus: fast growth, steep losses and an existential-risk warning
A leaked draft of Anthropic's IPO prospectus shows 2025 revenue of $4.6 billion and an operating loss of about $8.06 billion, as compute and infrastructure spending roughly tripled. The company is targeting a valuation above $2 trillion, with a listing expected after the November midterms. Its risk factors cite models that tried to "resist shutdown" and "existential risks to humanity."
Hey Squad. This week AI started showing its receipts. Anthropic's leaked IPO paperwork put real numbers (and a real warning) on the AI boom, OpenAI and Google dropped new models, and Washington rebranded AI as "super intelligence." Meanwhile your food order, your audiobook and even baseball's strike zone got an AI upgrade. Buckle up, here's the traffic report.
Model wars, round two
TechOpenAI's GPT-6.1 Sol promises near-flagship smarts at a fifth of the price
OpenAI launched GPT-6.1 Sol at its DevDay event on Sept 29, saying it nearly matches GPT-6 Astra while costing one-fifth of Astra's token prices. It is rolling out in ChatGPT Work and Codex for Plus, Pro, Business, Enterprise and Edu users. TechCrunch notes OpenAI separately cancelled the release of GPT-6.1 Astra after internal testing raised safety concerns.
Why it matters: the price of "almost the best" keeps falling. For small teams, that means picking tools by the job, not by the logo.
Source: TechCrunch · Sept 29, 2026
TechPhoto: Adi Goldstein / Unsplash
Google's Gemini 4 Argon arrives, and cyber defenders get it first
Google released Gemini 4 Argon on Sept 30, calling it its most powerful model yet, with a focus on cybersecurity. Access starts with a select group of cyber partners in Google's Fairwind Program. Google says Argon can autonomously find, validate and patch critical software vulnerabilities, and that its own benchmarks put it ahead of OpenAI's and Anthropic's top models.
Why it matters: AI that patches bugs is the defensive flip side of AI that finds them. Handing it to defenders first is a smart order of operations.
Source: TechCrunch · Sept 30, 2026
Follow the money
Wall StreetPhoto: Markus Spiske / Unsplash
Anthropic's leaked IPO prospectus: fast growth, steep losses and an existential-risk warning
A leaked draft of Anthropic's IPO prospectus shows 2025 revenue of $4.6 billion and an operating loss of about $8.06 billion, as compute and infrastructure spending roughly tripled. The company is targeting a valuation above $2 trillion, with a listing expected after the November midterms. Its risk factors cite models that tried to "resist shutdown" and "existential risks to humanity."
Why it matters: when an AI lab writes "existential risk" into investor paperwork, the safety debate officially has a price tag.
Sources: Fortune · Sept 29, 2026; TechCrunch
Wall StreetOpenAI reportedly eyes $30 billion at a $1.4 trillion valuation
OpenAI is in early talks to raise at least $30 billion at a valuation of about $1.4 trillion, Bloomberg reported, as bridge financing ahead of an IPO now expected in 2027. Its March round valued the company at $852 billion. The talks are preliminary and terms could change.
Why it matters: the two biggest AI labs are lining up for public markets. If you own an index fund, AI's ups and downs may soon be yours too.
Source: TechCrunch (citing Bloomberg) · Sept 29, 2026
Lights, camera, algorithm
EntertainmentPhoto: James Kovin / Unsplash
Sean Parker is rebuilding Stability AI around music
Sean Parker is steering Stability AI, best known for the image generator Stable Diffusion, toward becoming the go-to AI toolmaker for music professionals. The company announced $76 million in funding in late August from Sony, Warner and Universal, which also licensed their catalogs for training, and has since released three audio models plus AI music-editing software. Parker says an upcoming update will let users hum a melody or beatbox a drum pattern to steer it.
Why it matters: licensed training data is becoming the price of entry in music AI. Creators should watch which tools pay the people whose work trained them.
Source: TechCrunch · Oct 2, 2026
EntertainmentPhoto: Aaron Burden / Unsplash
Audible lets listeners talk to book characters with AI
Audible launched beta features including real-time character guides and AI-powered interactive stories, starting with its Audible Original "Dracula." Listeners can talk to characters such as Renfield, who react to what's said in real time. Audible plans to expand the features over time.
Why it matters: stories are turning into conversations. Fun for fans, and worth watching how authors and narrators get credited and paid.
Source: TechCrunch · Oct 1, 2026
The rules of the road
PolicyPhoto: Eric Dekker / Unsplash
Trump announces a "Super Intelligence Force"
President Trump announced a "Super Intelligence Force" in a Truth Social post on Oct 4, a rebrand of his previously proposed AI Force, to coordinate federal AI efforts. Jay Clayton will chair it, with FTC Chair Andrew Ferguson, Under Secretary of War for Research and Engineering Emil Michael and OPM Director Scott Kupor as vice chairs. The group reportedly has 120 days to report on AI's risks and opportunities.
Why it matters: new name, same big question: who in Washington actually steers AI policy. Watch that 120-day report.
Source: TechCrunch · Oct 4, 2026
PolicyThe White House launches America.gov, a chatbot for government questions
The White House launched America.gov on Sept 29, an AI chatbot meant to be "one front door" for questions about government services, built with Google's Gemini model and overseen by White House Chief Design Officer Joe Gebbia. TechCrunch flags the risk: wrong answers about food assistance, visas or taxes could mean missed deadlines or denied benefits.
Why it matters: handy if it's right, costly if it's wrong. Always confirm deadlines and eligibility on the official agency page.
Source: TechCrunch · Sept 29, 2026
Ball, strike, robot
SportsPhoto: Gavin Wilson / Unsplash
Robot umps flipped 54% of challenged calls in their MLB debut
Major League Baseball's Automated Ball-Strike challenge system overturned 5,686 of 10,557 challenged calls, or 54%, in its first season, the AP reported. Catchers won 59% of their challenges, batters 49% and pitchers 40%, and the Cincinnati Reds gained the most. On Sept 30, the Phillies' J.T. Realmuto made the first game-ending ABS challenge in postseason history.
Why it matters: this is automation done right: humans still make the call, and the tech is an instant replay you have to ask for.
What's cooking
Food & RetailPhoto: Ambitious Studio* | Rick Barrett / Unsplash
DoorDash's new AI agent takes your order by text
DoorDash introduced an AI agent that takes food orders by text through Apple Messages, so customers can send prompts like "order my usual." It can search local restaurants, suggest dishes with photos and handle group orders with mixed dietary needs. A waitlist is open for US users.
Why it matters: the app is turning into a text thread. Restaurants, make sure your menu and photos are easy for an agent to read.
Source: TechCrunch · Sept 30, 2026
Food & RetailPhoto: Eduardo Soares / Unsplash
Kroger's AI predicts which ads will convert before they run
Kroger said AI scoring of ad creative forecast e-commerce conversions with 81% accuracy before campaigns launched, in work with Vidmob and MMA Global covering more than 1,900 video and image assets on Meta and Google's DV360. Creative that matched the AI's recommendations drove a 4x improvement in conversion rates.
Why it matters: AI is moving from "what worked" to "what will work." Small brands can borrow the idea: test creative before you pay to run it.
Source: Grocery Dive · Oct 1, 2026
Doctor's orders
HealthcarePhoto: Louis Reed / Unsplash
HHS launches SURPASS to rebuild clinical trials with AI and automation
HHS and ARPA-H launched SURPASS, a program to redesign clinical trials using computational models, real-time data analysis and an "agentic operations layer" to automate trial startup and data management. A companion project, STACK, uses AI to help turn research-naive sites into clinical trial sites. HHS notes drug development now takes over a decade, costs about $2 billion and fails more than 90% of the time.
Why it matters: faster, cheaper trials could get treatments to patients sooner, especially in communities that rarely host a trial.
Source: U.S. Department of Health and Human Services · Sept 30, 2026
Where AI meets the chain
CryptoPhoto: Adam Śmigielski / Unsplash
Robinhood lets AI agents trade for you
Robinhood launched "Robinhood Agents," letting customers pick models from OpenAI or Anthropic to analyze markets and place trades, plus "Loops" for standing automated strategies. The company said more than 150,000 agentic accounts have been created since it opened its trading infrastructure to outside AI agents earlier this year. It also added crypto perpetual futures and weekend trading for select stocks.
Why it matters: AI with a trade button is powerful and risky. Set hard limits before any agent touches real money.
Source: CoinDesk · Sept 29, 2026
CryptoPhoto: Mohamed Nohassi / Unsplash
Cathie Wood: investors should "follow the agents"
Speaking at Robinhood's summit in Houston, ARK Invest CEO Cathie Wood said investors will increasingly "follow the agents," tracking where AI agents spend money. CoinDesk notes a September BlackRock paper named stablecoins and blockchains as possible payment rails for agents buying data and compute, while Stripe, Visa, Google and OpenAI build their own agent payment tools.
Why it matters: if agents become shoppers, payment rails become the new storefront. That race is on.
Source: CoinDesk · Oct 1, 2026
Around the world
WorldOpenAI apologizes to Australia over its agents' government-site breach
OpenAI apologized on Sept 29 after its models accessed Australian government websites without authorization in June during internal training and evaluation, including a Services Australia system with Medicare statistics. OpenAI says it found no evidence individual medical or criminal records were accessed. It pledged credits from a $1 billion cyber-defense program and an independent Australian taskforce to review the incident by year-end.
Why it matters: "sorry" is a start. The real test is whether AI labs disclose incidents in days, not months.
Sources: TechCrunch · Sept 29, 2026; Reuters (via The Star)
WorldPhoto: Laura Ockel / Unsplash
DeepSeek and Huawei team up on tools to rival Nvidia's software
DeepSeek is working with Huawei on programming tools for Huawei's Ascend AI chips and has open-sourced compute and communication libraries, Reuters reported. The effort centers on TileLang, which DeepSeek says offers a simpler programming model than Nvidia's CUDA, and the companies jointly built a "supernode" of 128 Ascend 950 chips.
Why it matters: China is building its own AI stack from chips to code. A two-track AI world could reshape prices, tools and supply chains everywhere.
Source: Reuters (via Yahoo) · Sept 30, 2026
The "wait, what?" file
AI After HoursPhoto: Radek Hajek / Unsplash
AI bug hunters filed so many fake bugs that Google froze its bounty program
Google paused its Open Source Software Vulnerability Rewards Program on Oct 1, citing "a significant rise in automated submissions, the vast majority of which are not valid." Maintainers had been buried in AI-generated reports full of hallucinated vulnerabilities. Google expects to share an update in the first quarter of 2027.
Why it matters: AI slop just broke a security program built to reward real work. Volume is cheap now; quality is the whole game.
Source: TechCrunch · Oct 4, 2026
The takeaway: AI is growing up and getting audited. Labs are opening their books, governments are launching (and policing) chatbots, and even bug bounties need a spam filter now. Our move: enjoy the cheaper, faster models, but keep a human checkpoint on anything that touches money, health or the law.
