Ten AI stories moved today. Four of them turn on one question. Who gets to run the shelf where everyone else's AI work sits?
The Justice Department asked that question about Nvidia first. Washington asked a version of it about Anthropic, too. One hand discounted Claude. The other hand blocked it.
Read the other eight stories against that split.
Washington Opens a File on Nvidia's Biggest Shelf
The Justice Department is examining Nvidia's $12.93 billion purchase of Hugging Face1. Developers use that site to share and find AI models.
Antitrust staff want to know one thing. Does owning that shelf let Nvidia squeeze rival chipmakers out of it2?
Two more Nvidia deals already sit under that same review. One is Run:ai. The other is SchedMD, plus a licensing arrangement with Groq2.
Jensen Huang, Nvidia's founder and chief executive, says the platform stays open regardless of who owns it. "Hugging Face will remain an open platform for the entire AI ecosystem," Huang said when the deal was announced. "Developers will choose the models they want, the frameworks they want, the clouds and inference service providers they want, and the computing platforms they want"1.
Jim Cramer reads the same deal differently. In September, the CNBC host said the market itself proves the point. "If AMD were announced, they were buying Hugging Face, if Broadcom announced they were buying Hugging Face, we would send NVIDIA down," Cramer said. "So you got to think of it like that. It's a little asymmetrical"3.
Anthropic's Filing Turns Revenue Into a Bigger Loss
Anthropic filed to list on the Nasdaq this week. Its ticker: ANT. Revenue for the first half of 2026 reached $16.3 billion, twelve times last year's figure4.
The net loss ran to $80.7 billion. Noncash charges tied to convertible debt drove most of that loss, far more than operating costs did4. A Founder LLC holding Anthropic's seven co-founders keeps 50.1 percent of the vote after the listing.
Anthropic Turns $16.3 Billion of Revenue Into an $80.7 Billion Loss
First half of 2026, billions of dollars, from Anthropic's S-1 filing
The first column is revenue, the magenta column is what costs and noncash charges took away, and the last column is where the half closed.
Source: TechCrunch [4]. Chart by Ryan Elliott Dennis.
The numbers behind this chart
| Item | Value |
|---|---|
| Revenue | $16.3B |
| Costs and charges | −$97B |
| Net loss | −$80.7B |
Spending tells a steadier story than the loss does. Anthropic's own reporting tools put net revenue retention at 900 percent, comparing the third quarter of 2025 with the third quarter of 20265. Spending on Claude Code has grown tenfold this year.
GSA Discounts Claude While the Pentagon Still Bans It
The General Services Administration widened its deal with Anthropic this week. Token prices for federal, state, local and tribal buyers dropped 40 percent6. Every agency pays only for what it uses, and the discount now covers Claude's government-only, FedRAMP-approved setup too.
Anthropic@AnthropicAIWe're expanding our partnership with @USGSA. Under a new OneGov agreement, we are bringing Claude to federal, state, local, tribal, and territorial governments at 40% off.https://t.co/p0lA6cP2y1
A federal court drew the opposite line in September. Judges upheld the Pentagon's designation of Anthropic as a supply-chain risk, by a vote of 2 to 17. The company had refused to let Claude run autonomous weapons or mass surveillance.
Gregory Katsas wrote the majority opinion. He said the Pentagon had reason to worry. "The Secretary raises the deeply sobering prospect of overly constrained AI models shutting down unexpectedly and thus causing important military operations to fail," Katsas wrote.
Karen Henderson dissented. She argued the law should credit Anthropic's own restraint, rather than punish it. The law, she wrote, "does not treat 'a contractor's honest and upfront enforcement of restrictions' as the kind of supply-chain risk that permits the government to blacklist it"8.
Anthropic asked the full appeals court to rehear the case on Oct. 13. That petition still waits on a ruling.
Google Sends Its Newest Model to Hunt Bugs First
Gemini 4 Carbon arrived this week. It was built for one job first: finding and fixing software flaws9. Google says the model scores 88.2 percent on SEC-bench Pro, a test of real vulnerability discovery. It scores 71 percent on CWE-bench, which checks whether a fix actually works.
Cyber defenders get Carbon first, through Google's Fairwind Program. Wider developer access follows later9. Koray Kavukcuoglu leads Google DeepMind's research as its chief AI architect, and he framed the sequencing as deliberate. "Pushing the frontier forward requires moving fast while remaining deliberate about deployment," Kavukcuoglu said.
Agents Now Draft Most of GitHub's Pull Requests
GitHub's yearly Octoverse report landed Wednesday. The number at its center is stark. More than 40 percent of pull requests in September started as an agent's own draft. In March, that share was 17 percent10.
AI Agents Draft 40 Percent of GitHub's Pull Requests, Up From 17 in March
Share of pull requests opened as an agent's first draft, GitHub's Octoverse report, March and September 2026
Each column is one month; the taller September column shows agents writing a larger share of pull requests first.
Source: The GitHub Blog [10]. Chart by Ryan Elliott Dennis.
The numbers behind this chart
| Item | Value |
|---|---|
| March | 17% |
| September | 40% |
Developers pushed code 17.9 billion times over the past year, more than 2024 and 2025 combined10. A repository that ships with its own agent configuration file logs eight times more commits than one that lacks it.
Microsoft Offers a Quarter of Its Staff the Door
Roughly 60,000 employees heard this week that they can take a voluntary buyout. That is about a quarter of Microsoft's workforce11. The offer reaches vice presidents and below in the United States, Canada, India and Britain. Data center, sales and marketing teams stay out of it.
Amy Coleman, Microsoft's chief people officer, framed the buyout as a choice rather than a cut. "The Voluntary Exit Offer reflects an important part of how we're evolving our approach to change: creating greater flexibility and, where possible, more ways for employees to shape their next chapter," Coleman said11. Microsoft keeps pouring money into AI data centers the same week.
OpenAI Reportedly Shops for Google's Chips
A report this week says OpenAI wants to buy Google's own AI chips outright. That goes past the renting it already does12. Both companies stayed quiet on the plan. Its price, volume and delivery date remain unknown.
OpenAI began renting Google's chips back in June 2025. A purchase would mark its clearest step yet away from Nvidia alone12.
Claude Moves Past Code Into Proof
Anthropic's safety researchers paired with the formal-methods firm Theorem this week. Together they pushed Claude past bug hunting13. The model wrote roughly 180 million lines of machine-checkable proof. It covers 239 entry points across 123 open-source libraries, pulled from Google's own fuzzing project.
The proofs show each program exits cleanly. Memory stays safe, for every input under 64 megabytes13. They cover one narrow setup only: static code, built for a single chip. Most of the libraries' real-world use sits outside what Claude proved.
Hackers Name the Staff Behind DHS's AI Tools
A leak this week put more than 41,000 DHS staff and contractors into public view. The leak grouped them by their internal teams14. Among those named were employees in groups tied to Palantir's software and to xAI's Grok account.
AI Lately covered the leak in full depth on its own page Wednesday. That report traces who the names belong to and what each access group controls.
A Lawsuit Blames ChatGPT for a Fatal Taper
OpenAI faces a lawsuit this week. A family says ChatGPT designed the medication schedule that preceded their daughter's death from epilepsy15. The complaint says the chatbot moved a 31-year-old off her seizure drug over 25 days. She followed that plan alone.
AI Lately's own report on the lawsuit ran Tuesday. It reads the filing against what OpenAI and outside legal scholars say about holding a chatbot liable for one person's death.
By the numbers
- $12.93 billion: the price Nvidia agreed to pay for Hugging Face, now under Justice Department review1.
- 40 percent: the discount the GSA added to Claude's government pricing this week6.
- 2 to 1: the vote by which an appeals panel kept Anthropic off Pentagon systems7.
- 88.2 percent: Gemini 4 Carbon's score on SEC-bench Pro, a real vulnerability-finding test9.
- 60,000: Microsoft staff offered a voluntary buyout this week, about a quarter of its workforce11.
- 180 million: the lines of machine-checkable proof Claude wrote to verify 123 open-source libraries13.
- More than 40 percent: the share of GitHub pull requests an agent drafted first in September, up from 17 percent in March10.
What to watch
Watch the D.C. Circuit's docket. Anthropic's rehearing petition, filed Oct. 13, still waits there. A reversal would let the Pentagon's biggest AI holdout back into the building it has been locked out of since February.
The DOJ's Nvidia review deserves the same attention. Watch for it to reach the Hugging Face deal directly, past the pattern of purchases around it.
Two more threads run today in AI Lately's columns. One asks whether you can keep your own chats out of a company's AI training. The other prices what Hollywood's new actors' contract puts on an AI performer's voice.
Sources
- "DOJ Probes Nvidia's Hugging Face Deal Over AI Competition Concerns," PYMNTS, Oct. 28, 2026, https://www.pymnts.com/legal/2026/doj-probes-nvidias-hugging-face-deal-over-ai-competition-concerns/
- "Nvidia faces DOJ antitrust scrutiny over $12.9B Hugging Face deal, Reuters says," Reuters, via Yahoo Finance, Oct. 28, 2026, https://finance.yahoo.com/technology/ai/articles/doj-inquires-nvidia-hugging-face-193103860.html
- "Cramer Says Nvidia Spent $13 Billion to Make Sure You Never Buy a Rival's Chip," 24/7 Wall St., Sept. 4, 2026, https://247wallst.com/investing/2026/09/04/cramer-says-nvidia-spent-13-billion-to-make-sure-you-never-buy-a-rivals-chip/
- "Anthropic officially files for its public debut," TechCrunch, Oct. 26, 2026, https://techcrunch.com/2026/10/26/anthropic-officially-files-for-its-public-debut/
- "Anthropic just launched tools to show companies what their AI spending is buying," TechCrunch, Oct. 28, 2026, https://techcrunch.com/2026/10/28/anthropic-just-launched-tools-to-show-companies-what-their-ai-spending-is-buying
- "GSA, Anthropic Reach OneGov Agreement on Discounted, Consumption-Based Pricing for Claude," U.S. General Services Administration, Oct. 27, 2026, https://www.gsa.gov/about-gsa/newsroom/news-releases/gsa-anthropic-reach-onegov-agreement-on-discounted-consumptionbased-pricing-10272026
- "US appeals court upholds Pentagon's blacklisting of Anthropic," Military Times, Sept. 25, 2026, https://www.militarytimes.com/news/pentagon-congress/2026/09/25/us-appeals-court-upholds-pentagons-blacklisting-of-anthropic/
- "Federal appeals court upholds Pentagon designation of Anthropic as supply chain risk," ABC News, Sept. 25, 2026, https://abcnews.com/Business/anthropic-appeals-court-declines-block-pentagon-blacklisting/story?id=136755690
- "Gemini 4 Carbon: our next frontier model," Google, Oct. 28, 2026, https://blog.google/innovation-and-ai/models-and-research/gemini-models/introducing-gemini-4-carbon/
- "Octoverse: Commits, pushes, reach new highs in 2026 as developers build with agents," The GitHub Blog, Oct. 28, 2026, https://github.blog/news-insights/octoverse/octoverse-commits-pushes-reach-new-highs-in-2026-as-developers-build-with-agents/
- "Microsoft offers voluntary exit packages to 60,000 staff as AI spending surges," Malay Mail, Oct. 27, 2026, https://www.malaymail.com/news/money/2026/10/27/microsoft-offers-voluntary-exit-packages-to-60000-staff-as-ai-spending-surges/240732
- "OpenAI reportedly plans to buy Google's TPU chips alongside Nvidia," The Next Web, Oct. 28, 2026, https://thenextweb.com/news/openai-google-tpu-chips
- "Claude Writes 180 Million Lines of Proofs to Verify 123 Open Source Libraries," DevOps.com, Oct. 28, 2026, https://devops.com/claude-writes-180-million-lines-of-proofs-to-verify-123-open-source-libraries
- "Hackers expose the identities of more than 40,000 DHS employees," NPR, Oct. 28, 2026, https://www.npr.org/2026/10/28/nx-s1-6021497/dhs-hack-ice
- "Parents sue OpenAI, alleging GPT-4o coached their daughter off seizure medication," TechCrunch, Oct. 27, 2026, https://techcrunch.com/2026/10/27/parents-sue-openai-alleging-gpt-4o-coached-their-daughter-off-seizure-medication/