Anthropic hired 33 people this year to watch for spies inside its own walls1. The count is more than double last year's. OpenAI's own team jumped to 22 from nine2.
Both buildups trace back to one scandal. It cost Nvidia its chief executive, and it started with a bug hidden behind a television at OpenAI's own office.
The Spy Hunt Grows at Anthropic and OpenAI
Reuters traces Anthropic's counterintelligence staff from nine people in 2024 to 33 this year1. It climbed through 14 last year.
Jensen Huang resigned as Nvidia's chief executive this month. Reports say a woman he dated was feeding information to Beijing. China denies the claim2. Prosecutors indicted an OpenAI engineer a month earlier for hiding a listening device behind a television in a meeting room.
Fortune reports that OpenAI's own security team jumped to 22 workers this year from nine in 20252. Nvidia added 16 more of its own.
Pay for the work now runs $200,000 to $500,000 a year, Revelio Labs data shows. The money draws former CIA and NSA officers into corporate badges.
Anthropic's Spy-Hunting Staff More Than Doubles to 33
Counterintelligence hires at Anthropic and OpenAI, 2025 versus 2026, from Revelio Labs hiring data
Each pair of columns is one company. The taller magenta column is this year, next to last year's shorter blue column.
Sources: Reuters [1]; Fortune [2]. Chart by Ryan Elliott Dennis.
The numbers behind this chart
| Item | 2025 | 2026 |
|---|---|---|
| Anthropic | 14 | 33 |
| OpenAI | 9 | 22 |
Catherine Marinis-Yaqub left the CIA for Control Risks, a security firm with AI clients now. She watches for two risks most: an employee copying trade secrets, and an executive open to blackmail1.
Kathy Lavinder places security chiefs for a living, and the pay climbs with the fear. Top corporate security chiefs now earn $1 million to $4 million a year, mostly in stock, she says.
Christopher Falkenberg built his career at the Secret Service. He later ran Insite Risk Management. "There are some very sophisticated people out of government thinking about security for these businesses," he told Reuters1. His read is simple: hiring still lags the industry's own demand.
Eric O'Neill spent years chasing spies for the FBI before Fortune asked him about AI's newest hires. "By the time a security professional raises concerns, the executive may already be emotionally invested and unwilling to listen," he said2. Huang's own exit ran through exactly that gap. An executive still decides whom to trust, whatever badge checks the lobby door.

Wall Street Prices Anthropic's Debut
Bankers set a $70 to $78 range for Anthropic's IPO on Nov. 103. The sale covers 825 million shares. At the top of that range, the sale could raise $64.4 billion.
CNBC expects pricing Nov. 18. Trading starts Nov. 19, under the ticker ANT. The filing already trimmed figures that had circulated a day earlier. A roadshow number keeps moving until a deal actually prices.
Kress Faces Her First Earnings Call Alone
Nvidia reports earnings Nov. 17, its first since Jensen Huang's exit4. Colette Kress now holds two jobs. She is both chief financial officer and interim chief executive.
Wall Street expects revenue near $109 billion. That is up roughly 91 percent from a year earlier. Nvidia's stock has fallen more than 30 percent since October.
Kress already told investors in August that tight supply held back faster growth. Analysts want to know whether supply still falls short of demand.
Investors Pour Money Into LangChain
LangChain raised $300 million this week at a $3 billion valuation5. CapitalG, Alphabet's growth fund, led the round. More than 7,000 groups now run agents through its LangSmith platform. The company says so.
Harrison Chase, the company's co-founder and chief executive, built the open-source code the business grew from. The new valuation more than doubles what investors paid a year ago, LangChain's last raise.
Enterprises Can Run Claude at Home Now
Anthropic opened a private beta letting companies run its agents inside their own cloud6. The self-hosted version keeps the harness, the sandboxes and session logs on a company's own machines. Model requests still travel out to Anthropic, Amazon Bedrock or another approved provider.
A company still asks its Anthropic account team for access. The beta stays invitation-only for now.
ClaudeDevs@ClaudeDevsSelf-hosted Claude Managed Agents is now in private beta.
You can host the agent harness, sandboxes, and session history, in your own cloud or on-premises, and choose where requests to Claude are sent. pic.twitter.com/JaiwF1SvKE
Outside Agents Get Grounded at Checkout
United Airlines is restricting outside AI agents from booking its flights7. The airline says the tools can trigger unintended purchases or burn through MileagePlus miles. Automated bookings might also break federal flight rules, United says.
A traveler's own agent can still search United's fares, though United finishes the purchase itself. Delta and American let some outside AI tools complete a booking, so the industry splits on the same question.
Ozzy Fox Comes From a Studio That Once Sued AI
Disney will bring "Ozzy Fox," an AI-animated series, to Disney+ in December8. French startup Animaj built it. Its tools turn artists' sketches and text into three-dimensional animation. Chief executive Josh D'Amaro now pushes AI into finance, streaming recommendations and the parks too.
The company sued Midjourney over its characters last year, and that case is still open. Disney also ended an earlier deal letting OpenAI's Sora app use its characters. The reversal came the same week D'Amaro took the top job.

Six Thousand March Toward Holyrood
More than 6,000 people marched through Edinburgh on Nov. 14, heading for the Scottish Parliament9. Kim Pratt led the campaign for Friends of the Earth Scotland. "Thousands came out to demand that the people of Scotland can have their say on these monstrous AI data centres," she said. Campaigners from towns across Scotland, from Cockenzie to North Ayrshire, traveled in to join the march.

Twenty-three proposed data centers would add 9.1 gigawatts to Scotland's grid, nearly tripling today's 5-gigawatt peak demand9. Scotland's government says it will publish new planning guidance on data centers before the year ends.
23 Proposed Data Centers Would Add 9.1 Gigawatts to Scotland's Grid
Scotland's electricity demand, current peak versus 23 proposed AI data centers, in gigawatts
The first column is today's peak demand. The magenta column is what the 23 proposed data centers would add. The last column is where demand would land if every one is built.
This counts every proposed data center as built. Friends of the Earth Scotland expects some applications to be rejected.
Source: Friends of the Earth Scotland [9]. Chart by Ryan Elliott Dennis.
The numbers behind this chart
| Item | Value |
|---|---|
| Current peak demand | 5GW |
| New centers add | +9.1GW |
| Total if all built | 14.1GW |
Luke Corbin's Agents Now Run AT&T's Network
AT&T now runs nearly 500 AI workflows in production10. It plans to triple that number next year. Those agents already handle more than 1 million transactions a month.
Luke Corbin, an AT&T vice president, says the work means the company now processes 45 billion AI tokens daily. His team redesigns whole processes, he says. Older efforts just automated single tasks alone. Andy Markus, the company's data chief, calls it a move from an information economy to an action economy.
DeepMind Proves Signal's Code Line by Line
Google DeepMind proved that Signal's own running code holds its security promises. Earlier proofs in this field covered only a system's design on paper, DeepMind says11. They stopped short of the actual code a phone runs.
Agents built on its Gemini model wrote more than six million lines of checked proof. The work took four weeks. It covers the encryption behind Signal and Google Messages.
Human experts still wrote every specification the agents proved against. Researchers even put back two bugs the team had already fixed, and the agents caught both before calling the code clean.
What to watch
Nvidia's earnings call on Nov. 17 should show whether Kress names a timeline for a permanent successor. Anthropic's roadshow still has to hold its $70 to $78 range through pricing on Nov. 18. A spy-hunting hire faces one real test: proof that a hire caught something only an insider could see.
Sources
- Toby Sterling, "Help wanted: AI firms court ex-spies as Nvidia drama spotlights security risks," Reuters, Nov. 12, 2026, https://www.1049thewolf.com/2026/11/12/help-wanted-ai-firms-court-ex-spies-as-nvidia-drama-spotlights-security-risks/
- Jason Ma, "The hottest job in AI right now may be counterintelligence as companies pay former spooks $500,000 to protect trade secrets," Fortune, Nov. 14, 2026, https://fortune.com/2026/11/14/ai-industrial-espionage-counterintelligence-jobs-china-spying-nvidia-jensen-huang-pentagon-economic-warfare/
- Tobi Opeyemi Amure, "Anthropic IPO Date: $70-$78 Price Range Set, Trading Reported From 19 November," FinanceFeeds, Nov. 11, 2026, https://financefeeds.com/anthropic-ipo-date-price-range-70-78-trading-19-november/
- Steven Porrello, "Nvidia Reports Earnings on Nov. 17. Here's What Investors Should Be Watching.," The Motley Fool, Nov. 15, 2026, https://www.fool.com/investing/2026/11/15/nvidia-reports-earnings-on-nov-17-2026-heres-what-investors-should-be-watching/
- Diego Almada Lopez, "LangChain raises $300 million at a $3 billion valuation as LangSmith passes 7,000 organizations," Crypto Briefing, Nov. 13, 2026, https://cryptobriefing.com/langchain-raises-300-million-3-billion-valuation/
- Anthropic, "Operate Claude Managed Agents inside your security perimeter," Anthropic, Nov. 12, 2026, https://claude.com/resources/articles/self-hosted-claude-managed-agents
- Cris Tolomia, "United Airlines is blocking personal AI agents from booking flights," Quartz, Nov. 11, 2026, https://qz.com/united-airlines-blocking-personal-ai-agents-booking-flights-111126
- Ana Maria Constantin, "Disney is now embracing the AI it once fought, from cartoons to finance," The Next Web, Nov. 13, 2026, https://thenextweb.com/news/disney-embraces-ai-ozzy-fox-damaro
- Friends of the Earth Scotland, "Thousands march to stop AI data centres," Friends of the Earth Scotland, Nov. 14, 2026, https://foe.scot/press-release/thousands-march-to-stop-ai-data-centres/
- Paul Gillin, "AT&T uses agentic AI to reinvent business processes from the ground up," SiliconANGLE, Nov. 14, 2026, https://siliconangle.com/2026/11/14/att-uses-agentic-ai-to-reinvent-business-processes-from-the-ground-up/
- Pushmeet Kohli and Swarat Chaudhuri, "Introducing AlphaProof Code: a breakthrough in verifying complex software at depth," Google DeepMind, Nov. 10, 2026, https://deepmind.google/blog/introducing-alphaproof-code/




