Three stacked metal film cans and a brass loupe on a dark walnut desk, lamp-lit still-life. No people, no logos.
NewsTech

Nexus put more than 153 million drivers-license scans up for sale. ChatGPT for Healthcare can read Epic charts, not write them.

Aaron Harme

A dark-web shop called Nexus claimed more than 153 million North American drivers-license scans. Brian Krebs traced the images toward a Louisiana identity-verification vendor. The FBI’s New Orleans office opened an inquiry. The shop’s login page now reads that the service is no longer available. 153 million-plus is Nexus’s claim.

ChatGPT for Healthcare can pull authorized Epic context, read-only. Anthropic will keep Fable 5.1 enterprise monitoring data in the customer’s own cloud later this fall. Sources put Anthropic in a $35 billion Lambda deal on a Hut 8 Texas campus Nvidia would lease. Axios says OpenAI and Anthropic are selling growth and caution in the same week.

In this Brief

  • Nexus: 153M+ drivers-license scans (Nexus’s claim); FBI New Orleans inquiry into an apparent idscan.net breach; site vanished
  • ChatGPT for Healthcare: read-only Epic; UCSF pilot; 325M is TechCrunch; 99.1% is OpenAI’s eval
  • Fable 5.1 / EFS: customer-held logs later this fall; Mythos trusted-access only
  • Lambda $35B: sources; Nvidia would hold a Hut 8 Texas lease; Anthropic declined
  • OpenAI + Anthropic IPOs: dual-messaging; prospectus “as soon as next week” is Axios, not a filing

The copies were already out.

Setup: Brian Krebs at Krebs on Security reports a new identity-theft service on the dark web, Nexus, is selling digital scans of more than 153 million drivers licenses from people in the United States and Canada. The New Orleans field office of the FBI launched an official inquiry into the source of the images the day Krebs published, September 1.

Nexus also claims more than 10 million identification cards, more than 3 million travel documents and/or international IDs, and at least 579,000 medical cards. Krebs writes the images appear to be siphoned from a widely used Louisiana identity-verification company, New Orleans-based idscan.net. The company told Krebs it was investigating and was not able to share additional information yet.

Nexus advertised on the Russian cybercrime forum Exploit. Some scans include infrared and ultraviolet pairs with timestamps. Krebs matched those stamps to rental-car and dispensary handoffs, including Hertz and a Planet13 store that idscan.net had an exclusive verification deal with.

Shortly after the story published, the Nexus site vanished. The login page was replaced with plain text: “This service is no longer available.”

Take: The scans are copies of IDs people handed over at rental-car counters and dispensary doors so a vendor could confirm the person was the person. The shop is selling that handoff.

I keep 153 million-plus on Nexus’s claim. Attribution to idscan.net is Krebs’s investigation plus an FBI inquiry into an apparent breach. The company has not fully admitted a breach in the story we opened.

The site is already gone. The copies were already out.

The chart stays the hospital’s.

Setup: OpenAI said Tuesday that healthcare organizations can bring authorized patient context from Epic into ChatGPT for Healthcare. The integration is read-only and does not write back to the chart. UCSF Health is the pilot partner.

A Healthcare Public Data plugin sits beside it, structured access to nine official sources including PubMed, DailyMed, and CMS Coverage.

Ivan Mehta at TechCrunch writes that Epic holds data for over 325 million patients. That figure is TechCrunch. The OpenAI post does not print it.

OpenAI’s EHR-context eval put 4,363 ratings across 27 clinical use cases, with physicians rating 99.1 percent of responses safe. ChatGPT for Healthcare customers can have an admin enable the EHR integration. It is not available on individual ChatGPT for Clinicians accounts.

Take: The model is a reader sitting on the hospital’s file.

A clinician can ask what changed since last visit without assembling the chart by hand. That is the pitch.

325 million stays on TechCrunch. 99.1 percent stays on OpenAI’s physicians rating OpenAI. Neither is a regulator’s audit.

UCSF is a pilot, not a hospital census. The EHR hook is an admin switch for ChatGPT for Healthcare customers. Individual clinicians do not get it.

The chart stays the hospital’s.

Longer cut: OpenAI connects Epic records and nine public healthcare datasets to ChatGPT for Healthcare.

The logs sit in a bucket the customer already pays for.

Setup: Anthropic announced Tuesday Enterprise Frontier Safeguards, a way to store monitoring data in the customer’s own Amazon S3, Azure Blob, or Google Cloud Storage, under keys the customer holds. Customer employees review flags by default. The product rolls out in phases later this fall, eligible customers keep zero data retention on Fable 5 and Fable 5.1 until then, and Anthropic does not charge for EFS while the vendor bills cloud storage.

The same day’s Fable 5.1 post made Claude Fable 5.1 generally available on AWS, Google Cloud, and Microsoft Azure. Mythos 5.1 is the same model with more permissive safeguards for vetted cybersecurity and life-science work, trusted-access only. Anthropic says it built EFS with more than 100 customers plus the three clouds, including a quarter of the Fortune 100 and every U.S. G-SIB.

Take: Enterprises wanted zero data retention. Anthropic wanted a rolling window so it could still catch misuse that spans sessions. EFS is the compromise: the search runs on logs in a bucket the customer already pays for.

Zero data retention until then is the bridge, not the product. The 30-day window is for correlating misuse across sessions and accounts, not for training. Anthropic says it has never trained on enterprise data without explicit permission, and never will.

Same weights. Different rooms. The logs are the enterprise story.

Longer cut: Anthropic will keep Fable 5.1 enterprise monitoring data in the customer’s own cloud later this fall.

The offtaker has not said yes on letterhead.

Setup: Reuters, reprinted by Channel NewsAsia, said a source familiar with the matter put Anthropic in a $35 billion cloud-computing deal with Nvidia-backed Lambda for a Texas data center in Nueces County of about 350 megawatts. The source spoke Monday on condition of anonymity because the information is not public. Anthropic, Nvidia, Hut 8, and Lambda did not immediately respond.

Shaurya Malwa at CoinDesk writes some of that capacity would run through Hut 8’s Beacon Point campus, with Nvidia holding the lease, Lambda using Nvidia chips, and Anthropic buying the compute. A Bloomberg reprint at The Business Times says a representative for Anthropic declined to comment.

Hut 8’s July 20 exhibit is the on-the-record floor: two 15-year leases, 352 megawatts a phase, 704 megawatts of IT capacity, $19.6 billion contracted value including a 3 percent annual rent escalator, same unnamed high-investment-grade tenant. CoinDesk asked Hut 8 how much of the 704 megawatts is this deal, and whether Nvidia is the undisclosed tenant. Hut 8 has not confirmed.

Take: Hut 8 puts up the hall, and Nvidia, on the sourced telling, would hold the lease and fill it. Lambda would sell the compute Anthropic would run. That is three Nvidia touchpoints and still nobody on the letterhead.

I keep the $35 billion on the sources line. About 350 megawatts is Reuters, not Hut 8’s filing, and the $19.6 billion is rent already on an exhibit to a tenant Hut 8 still will not name. Those dollars do not add.

CNA notes Anthropic said last week it will spend $45 billion to rent capacity from Nscale in West Virginia. That is a third contract, a different counterparty, a different state.

The offtaker has not said yes in public.

Longer cut: Sources put Anthropic in a $35 billion Lambda cloud deal on a Hut 8 Texas campus Nvidia would lease.

The two pitches will keep swapping rooms.

Setup: Ina Fried at Axios writes this morning that OpenAI and Anthropic are trying to convince Wall Street their businesses are sound and fast-growing while assuring governments their models do not pose unacceptable risks. Both are aiming at potentially record-breaking IPOs soon.

Axios says Anthropic could file a publicly available prospectus as soon as next week. OpenAI is in earlier stages of its IPO process. That timetable is Axios reporting, not a filing.

Anthropic is striking a commercially friendly note. OpenAI is more sober: Astra’s most powerful cyber capabilities start with trusted testers, and the company warns safeguards may slow, pause, or stop legitimate work.

Take: Wall Street wants a growth machine. Washington wants a cautious lab. Axios says both rooms are getting a different pitch in the same week.

Prospectus “as soon as next week” is Axios, not a document we opened. Neither listing is priced or filed on the pages we opened.

Dean Ball, OpenAI’s head of strategic futures, wrote that the Hugging Face incident is likely only the beginning of systems slipping human containment, including agents that pay their own compute bills. That essay is Axios’s intrigue line, not a prospectus risk factor.

The two pitches will keep swapping rooms.

Also noted

Maria Curi at Axios reports from Chapel Hill that Commerce and the White House Office of Science and Technology Policy are jockeying over the AI agenda at this week’s G20 Innovation Ministerial. A White House spokesperson told Axios that is fake news and that Commerce and OSTP are jointly leading a productive summit. Axios says the White House has finalized but not publicly released a framework for reviewing some frontier models before deployment.

The Wall Street Journal reports Elon Musk wants to make power-turbine components.

Tuesday’s Brief: the FTC says Amazon hid a surcharge so ad winners paid their own bid. Nvidia invested $3.5 billion in MediaTek convertibles.

Sources

One thought on “Nexus put more than 153 million drivers-license scans up for sale. ChatGPT for Healthcare can read Epic charts, not write them.”

Leave a Reply

Your email address will not be published. Required fields are marked *