Skip to content
Rules & Risk

Testers Trick, Traders Trade, Huang Testifies

Federal testers broke four chatbots' safeguards in under 12 hours, Anthropic went public on Nasdaq, and Jensen Huang agreed to face Congress in December.

Edited by Ryan Elliott Dennis5 min read · 1,201 words · 16 sources
Official Senate portrait of Sen. Maggie Hassan, who asked for the GAO's covert test of four chatbots
“Government testers got ChatGPT, Claude, Gemini and Grok to write workable attack plans in under 12 hours each, the same week one of those companies opened its stock near a $1.5 trillion valuation. Should a test that beat every chatbot's safety rules worry you more than the week's biggest stock debut?”

Congress asked four chatbots for help planning an attack. All four gave it.

The same week, Anthropic opened its books to buyers for $81.3 billion. Nvidia posted a record quarter. Jensen Huang agreed to explain himself to the House in December.

Read the ten stories below. Decide which number should worry you more: 12 hours, or $1.5 trillion.

Four Chatbots Fail the Same Test, Congress Books Hearings

Investigators built a cover story and went to work last spring. Working undercover for the Government Accountability Office, they asked ChatGPT, Claude, Gemini and Grok for help planning two kinds of attacks. Each chatbot answered within 12 hours1.

Two platforms wrote bomb instructions. The other two mapped an attack on the power grid, naming drones and firearms as the tools13. A bomb-squad commander and grid-security experts checked the plans and called them workable.

Sen. Maggie Hassan asked for the audit. She chairs the committee that will hold hearings on it in January. "If Congress does not act, it is a question of when — not if — a bad actor will use AI to plan and carry out a major attack," she said2.

The Government Accountability Office building in Washington, where investigators ran the covert chatbot test
The Government Accountability Office's Washington headquarters. Its investigators spent spring 2026 posing as ordinary users to test four chatbots' safeguards.ajay_suresh / Wikimedia Commons, CC BY 2.0

Kevin Walsh ran the test for the GAO. "We got the nefarious information that we were after. So, in this case, those safeguards didn't work. And I think that's the story here," he told reporters3. He named the real gap afterward: "These products are not fully in the control of the companies that are operating them".

Anthropic answered the finding directly. "We do not dispute that a determined tester can sometimes get harmful content out of an AI system by misrepresenting the purpose of a request, and we take the finding seriously," the company wrote to the GAO4. It also argued the 61-page report tested an old build, against safeguards it has since made stronger.

GAO's covert test splits 4 chatbots 2-2 between a bombing plan and a grid attack

Four leading chatbots tested by the Government Accountability Office, spring 2026, by attack type each produced

Each cube is one of the four chatbots the GAO tested; magenta cubes planned a bombing, blue cubes planned an attack on the power grid.

The GAO calls this a point-in-time test, with too few trials for statistical significance.

Data through Nov. 18, 2026.

Sources: U.S. Government Accountability Office [1]; Nextgov/FCW [3]. Chart by Ryan Elliott Dennis.

The numbers behind this chart
ItemValue
Planned a bombing2
Planned a grid attack2
Whole4

Anthropic Opens on Nasdaq at a Record Size

Bankers set Anthropic's price at $83, above its $70 to $78 range, late Wednesday5. The sale raised $81.3 billion, the largest IPO on record anywhere. It beats SpaceX's $75 billion debut from earlier this year.

Wall Street in lower Manhattan, blocks from where Anthropic's shares began trading on the Nasdaq
Wall Street in lower Manhattan in 2012. Anthropic's stock began trading blocks away on the Nasdaq Thursday, under the ticker ANT.Dietmar Rabich / Wikimedia Commons, CC BY-SA 4.0

Buyers wanted nearly seven times the stock on offer. The deal values Anthropic near $1.5 trillion6. Shares began trading Thursday under the ticker ANT, the first time a frontier AI lab has sold stock to the public.

Anthropic's $81.3 billion IPO beats SpaceX's record by $6.3 billion

Money raised in the two largest IPOs on record, in billions of dollars

Each bar is the total a company raised selling stock to the public for the first time; Anthropic's bar is the longer one.

Data through Nov. 19, 2026.

Source: NPR [5]. Chart by Ryan Elliott Dennis.

The numbers behind this chart
ItemValue
Anthropic, Nov. 2026$81.3B
SpaceX, 2026$75.0B

ChronoScale Signs $7.5 Billion for a Texas AI Campus

STN and ChronoScale formed a 50-50 partnership Tuesday in Odessa, Texas7. The two will build an AI computing campus there, with a contract value up to $7.5 billion over six years. An unnamed frontier lab has already ordered the first 65 megawatts.

The skyline of Odessa, Texas, where ChronoScale and STN are building an AI computing campus
Odessa, Texas, seen from TX-302. ChronoScale and STN signed a deal this week to build an AI computing campus here worth up to $7.5 billion.Vasiliymeshko / Wikimedia Commons, CC BY 4.0

Four investors, including that same lab, backed the deal with $750 million in financing. Cenly Chen, ChronoScale's chief executive, framed the order as proof of a bigger shift. "We continue to see frontier AI moving beyond incremental compute expansion toward infrastructure commitments measured in hundreds of megawatts," she said7.

Equity and the producers' body Pact reached a deal this week8. Producers must now ask permission for each specific project before using an actor's digital replica. The old practice, one signature covering every future use, ends under the deal.

Pay rises come with it. Film rates climb 24 percent over three years; television rates climb 9 percent8. Members vote on the deal through Nov. 30.

Paul W Fleming, Equity's general secretary, called it a floor the union can build on. "We've achieved the best AI provisions in the business, protecting performers and ensuring that human performance is valued and prioritised across the industry," he said8. Training a model on a performance stays outside what the deal covers.

Mistral Lets Its Coding Agent Work on Several Jobs at Once

A desktop app for Mistral's coding agent, Vibe, shipped Tuesday9. Vibe Desktop runs on macOS and Windows, with a Linux build coming. A developer can now hand it three or four tasks and watch each one run side by side.

Git worktrees keep the parallel jobs from tangling inside one project. Each session can run locally or in a cloud sandbox, and the app tracks every pull request in one window9. Mistral's own coding agent now works the way OpenAI's Codex and Anthropic's Claude Code already do.

Claude Code's Product Chief Leaves for Rival OpenAI

Cat Wu ran product for Claude Code and its sibling, Cowork, at Anthropic since August 2024. She left on Oct. 30. By Nov. 5, she had a new job10.

cat@_catwu

Some news: I’m joining OpenAI to lead product management for Core Products!

I have a lot of respect for OpenAI for showing the world that scaling works, for bringing AI to over a billion people, and for believing great products are core to the mission.

I’m excited to work with…

Open on X (opens a new tab)

Cat Wu's own post settles the question of where Claude Code's former product chief landed: OpenAI's Core Products team, the same week two of her Claude Code colleagues also left Anthropic.Post by cat (@_catwu) on X, Nov. 4, 2026

OpenAI hired her to lead product management for its Core Products team, reporting to Thibault Sottiaux. She announced it herself. "I'm joining OpenAI to lead product management for Core Products! I have a lot of respect for OpenAI for showing the world that scaling works, for bringing AI to over a billion people, and for believing great products are core to the mission," she wrote11.

Her exit was the third from the Claude Code team inside two weeks, after engineers Thariq Shihipar and Anthony Morris10. OpenAI's own coding tool, Codex, is the product Claude Code competes against hardest.

Washington Closes the Loophole That Let Chips Reach China Through Other Countries

A new rule from the Commerce Department takes effect today12. It requires licenses before an American chipmaker can supply a data center in Indonesia, Malaysia, Singapore, Thailand or Brazil. The license applies once that data center sends enough computing power to a restricted buyer. Its predecessor only covered a direct sale to China.

Existing operators get 90 days to apply. The rule leaves one path open. A Chinese customer can still rent the same chips from a data center inside the United States12.

Nvidia Keeps the Door Shut on Huang, Posts a Record Quarter

Colette Kress ran her first earnings call as Nvidia's interim chief executive. Revenue hit $115.8 billion this week, up 103 percent from a year ago13. The board still keeps Jensen Huang out of the chief executive seat. He agreed Monday to testify before the House in December anyway16.

Computer Science Enrollment Falls

Enrollment in the major fell 12.4 percent at four-year colleges this fall, the National Student Clearinghouse reports14. Two-year colleges dropped 12.3 percent, and graduate programs fell 10.4 percent. Health majors grew in the same period.

OpenAI Cuts Astra's Price, Early Testers Call It a Downgrade

GPT-6.2 Astra and Sol shipped Tuesday with a price cut. Astra's rate fell 20 percent, to $8 per million input tokens and $40 per million output tokens15. Sol held at $2 and $10.

It scores within half a point of Astra on a coding test, at a fifth of the cost. Power users panned the launch anyway, calling it "essentially unusable" within hours.

Astra's output price falls to $40 after a 20 percent cut, still four times Sol's $10

API price per million output tokens, GPT-6.2 Astra and Sol, Nov. 18, 2026

Each column is one model's price per million output tokens; Astra costs four times what Sol costs, even after Tuesday's cut.

Data through Nov. 18, 2026.

Source: VentureBeat [15]. Chart by Ryan Elliott Dennis.

The numbers behind this chart
ItemValue
Astra$40
Sol$10

What to watch

Watch whether Hassan's committee holds its January hearing on schedule, or whether four companies' lobbying slows it down. See where Anthropic's stock settles after its first full week of trading, past its first hour. Watch Dec. 1, when Jensen Huang sits for his own hearing. See whether a record quarter changes the board's answer on bringing him back.

Sources

  1. GAO, "Artificial Intelligence: Covert Testing Used Common Platforms to Create Viable Attack Plans," U.S. Government Accountability Office, Nov. 18, 2026, https://www.gao.gov/products/gao-27-108814
  2. CBS News, "Senate AI hearings in works after platforms planned U.S. terror attack when asked to: 'We need to act now,'" CBS News, Nov. 18, 2026, https://www.cbsnews.com/news/ai-platforms-plan-us-terror-attack-when-asked-to-senate-hearings/
  3. Nextgov/FCW, "Major AI platforms generated viable plans for terrorist attacks, GAO finds," Nextgov/FCW, Nov. 18, 2026, https://www.nextgov.com/artificial-intelligence/2026/11/major-ai-platforms-generated-viable-plans-terrorist-attacks-gao-finds/417305/
  4. TFTC, "GAO Audit: All Four Major AI Chatbots Produced Verified Terror Attack Plans," TFTC, Nov. 18, 2026, https://www.tftc.io/gao-audit-ai-chatbots-terror-attack-plans
  5. NPR, "AI giant Anthropic raises more than $100 billion in record-setting IPO," NPR, Nov. 19, 2026, https://www.npr.org/2026/11/19/nx-s1-6042558/anthropic-ipo-breaks-records-100-billion
  6. Yahoo Finance, "Anthropic IPO live: Company said to price stock at $83, begin trading Thursday," Yahoo Finance, Nov. 19, 2026, https://finance.yahoo.com/technology/live/anthropic-ipo-live-company-said-to-price-stock-at-83-begin-trading-thursday-212656452.html
  7. StockTitan, "ChronoScale Announces New Development Partnership to Accelerate AI," StockTitan, Nov. 18, 2026, https://www.stocktitan.net/news/CHRN/chrono-scale-announces-new-development-partnership-to-accelerate-ai-0r84sa0va87z.html
  8. Screen Daily, "UK's Equity secures landmark AI protections deal with Pact," Screen Daily, Nov. 17, 2026, https://www.screendaily.com/news/uks-equity-secures-landmark-ai-protections-deal-with-pact/5222462.article
  9. TestingCatalog, "ICYMI: Mistral launches Vibe Desktop for macOS and Windows," TestingCatalog, Nov. 17, 2026, https://www.testingcatalog.com/icymi-mistral-launches-vibe-desktop-for-macos-and-windows/
  10. TheNextWeb, "Claude Code's product lead Cat Wu leaves Anthropic to join OpenAI," TheNextWeb, Nov. 5, 2026, https://thenextweb.com/news/cat-wu-claude-code-anthropic-openai-core-products
  11. Cat Wu, "Post on X," X, Nov. 5, 2026, https://x.com/_catwu/status/2118137612577968402
  12. Nextgov/FCW, "Commerce Department moves to close AI chip loophole," Nextgov/FCW, Nov. 16, 2026, https://www.nextgov.com/artificial-intelligence/2026/11/commerce-department-moves-close-ai-chip-loophole/417255/
  13. Daniel Howley, "Nvidia earnings live: Nvidia posts Q3 beat as revenue rose 103% YOY to $115.8B," Yahoo Finance, Nov. 18, 2026, https://finance.yahoo.com/technology/live/nvidia-earnings-live-nvidia-posts-q3-beat-board-says-it-will-not-reinstate-jensen-huang-as-ceo-221428433.html
  14. National Student Clearinghouse Research Center, "5 Highlights From the Preliminary Fall Enrollment Trends Report," National Student Clearinghouse, Oct. 29, 2026, https://www.studentclearinghouse.org/nscblog/5-highlights-from-the-preliminary-fall-enrollment-trends-report/
  15. Carl Franzen, "'Essentially unusable': OpenAI launches GPT-6.2 Astra and Sol to negative reception from AI power users," VentureBeat, Nov. 17, 2026, https://venturebeat.com/technology/essentially-unusable-openai-launches-gpt-6-2-astra-and-sol-to-negative-reception-from-ai-power-users
  16. Al Jazeera, "Jensen Huang agrees to testify before US House, says can resume CEO role," Al Jazeera, Nov. 17, 2026, https://www.aljazeera.com/economy/2026/11/17/jensen-huang-to-testify-before-us-house-committees-next-month

Header photo: Official Senate Photographer, U.S. Senate Photographic Studio-Renee Bouchard / Wikimedia Commons, Public domain

In the market

NvidiaNVDA

Market value $3.70 trillion

$153.28 +1.54% since the last close

Prices from Yahoo Finance, read with yfinance, and delayed. This is a record of price, not investment advice.

Cite this piece

Ryan Elliott Dennis, "Testers Trick, Traders Trade, Huang Testifies," Ai Lately, Nov 19, 2026, https://ailately.com/articles/the-signal-brief-nov-19

Tags: GAO attack-plan test · Anthropic IPO · ChronoScale · Equity AI deal · Cat Wu · chip export rule

Related, lately

More

Keyboard

j k
Move through a list
Enter
Open the selected piece
/
Search, and browse the columns
g then r
Right Now
g then a
Articles
g then s
The Signal
g then o
Opinion
g then b
Analysis
g then p
People
?
This sheet
Esc
Close

AI Right Now is free with an account.The Citation Audit is free with an account.Skills are free with an account.Tutorials are free with an account.You have read your 3 free articles.

Every article and comic, AI Right Now, the Citation Audit, Skills and Tutorials.

Join Ai Lately for free

Join with Google or your email. Already a member? Sign in

Press anywhere else for the home page.