Latest / Neural Newscast / Why Lawmakers Want to Ban AI Health Data Sales [Model Behavior]
Transcript
- [00:00] Nina Park: From Neural Newscast, this is Model Behavior, AI-focused news and analysis on the models [00:05] Nina Park: shaping our world. [00:12] Nina Park: I'm Nina Park. Welcome to Model Behavior. This program provides an analytical look at [00:17] Nina Park: the mechanics of artificial intelligence, from the underlying infrastructure to the [00:24] Announcer: Today is June 29, 2026, and we are examining the tightening of medical privacy standards [00:31] Announcer: and a massive legal challenge from the local press. [00:34] Thatcher Collins: And I'm Thatcher Collins. [00:36] Thatcher Collins: We begin with a legislative proposal targeting the data brokers and developers who process [00:40] Thatcher Collins: sensitive personal information. [00:43] Thatcher Collins: Earlier today, Senator Elizabeth Warren and Representative Mary Gay Scamlin announced [00:47] Thatcher Collins: they are introducing an updated version of the Health and Location Data Protection Act. [00:52] Thatcher Collins: This bill is specifically engineered to address the data-intensive nature of modern generative AI models. [00:58] Announcer: The timing is notable, Thatcher. [01:00] Announcer: Over the past year, we have observed a rapid pivot by major AI firms into the healthcare sector. [01:06] Announcer: In January, OpenAI launched ChatGPT Health, encouraging users to upload medical records and MRI scans for analysis. [01:15] Announcer: Elon Musk's XAI made a similar push with Grok. [01:19] Announcer: Lawmakers are concerned that this intimate data could be leveraged or sold without the explicit consent required in clinical settings. [01:27] Thatcher Collins: Exactly. While companies like OpenAI and Anthropics state that their healthcare tiers are sandbox protected or HIPA ready, individual users often fall through the cracks of general privacy policies. [01:39] Thatcher Collins: Professor Sarah Gerke from the University of Illinois has highlighted that the United States lacks a federal framework to prevent the commercialization of this data. [01:46] Thatcher Collins: Okay. [01:47] Thatcher Collins: This new bill would authorize $1 billion for the FTC to enforce these protections over the next decade. [01:53] Announcer: Turning to the courts, a substantial legal challenge has emerged against OpenAI and Microsoft. [01:59] Announcer: On June 24th, a coalition representing nearly 400 local news organizations filed a lawsuit in Manhattan federal court. [02:07] Announcer: As reported by The Next Web, this is the largest copyright case brought by the local press to date. [02:12] Announcer: The litigation is led by Richner Communications, representing a diverse array of community-focused publications. [02:19] Thatcher Collins: What distinguishes this case, Nina, is the focus on regional reporting. [02:23] Thatcher Collins: We are talking about titles like the Arkansas Democrat Gazette and the New York Amsterdam News. [02:29] Thatcher Collins: The complaint alleges that OpenAI and Microsoft systematically crawled these sites, bypassing paywalls to ingest content for training— [02:38] Thatcher Collins: The plaintiffs argue that this unauthorized use of their intellectual property directly undermines the economic viability of local reporting. [02:46] Announcer: The publishers describe this as a death knell for investigative journalism. [02:51] Announcer: Their counsel Matthew Platkin points out that while a model can summarize a meeting, it cannot attend one or investigate local corruption. [02:59] Announcer: The lawsuit cites Sam Altman's own testimony before the House of Lords. [03:03] Announcer: In that testimony, Altman conceded that it would be impossible to train high-performing models without access to copyrighted material, which the plaintiffs view as an admission. [03:13] Thatcher Collins: The legal strategy relies on the Digital Millennium Copyright Act. [03:18] Thatcher Collins: The publishers argue that the AI firms did not just take the text, they stripped away copyright management information, like author bylines and terms of use. [03:27] Thatcher Collins: By removing these identifiers during ingestion, the models were allegedly able to memorize and reproduce reporting word for word, without providing any attribution or traffic back to the original local news sources. [03:40] Announcer: It is a high-stakes moment for OpenAI, especially as they move toward an estimated valuation of $852 billion. [03:48] Announcer: Meanwhile, the industry continues to advance. [03:51] Announcer: We recently saw Suno launch its Spark incubator for independent musicians. [03:56] Announcer: And in China, ZAI is claiming that its latest models match the cybersecurity safety benchmarks established by Anthropic. [04:04] Announcer: The dual pressure of product expansion and regulation is creating a very volatile environment. [04:10] Thatcher Collins: It underscores a fundamental tension—the valuation of these AI labs versus the cost of the data used to build them. [04:17] Thatcher Collins: Whether we are discussing health records or local obituaries, the era of harvesting free data for training appears to be closing. [04:26] Thatcher Collins: Lawmakers and creators are demanding a seat at the table and fair compensation for their intellectual property. [04:31] Thatcher Collins: We will be tracking the progress of that FTC funding quite closely, Nina. [04:36] SPEAKER_03: Thank you for listening to Model Behavior, a Neural Newscast editorial segment. [04:42] SPEAKER_03: For more details on these stories, visit us at mb.neuralnewscast.com. [04:48] SPEAKER_03: Neural Newscast is AI-assisted, human-reviewed. [04:52] SPEAKER_03: View our AI transparency policy at neuralnewscast.com. [04:56] Nina Park: This has been Model Behavior on Neural Newscast. [05:00] Nina Park: Examining the systems behind the story.