Neural Newscast

[00:00] Announcer: From Neural Newscast, this is Model Behavior, AI-focused news and analysis on the models
[00:05] Announcer: shaping our world.
[00:10] Nina Park: I'm Nina Park.
[00:13] Nina Park: Welcome to Model Behavior.
[00:14] Nina Park: It is March 25, 2026.
[00:17] Nina Park: Today, we're examining a critical legislative push to formalize AI safety boundaries and
[00:23] Nina Park: human-in-the-loop requirements within the United States military.
[00:27] Thatcher Collins: I'm Thatcher Collins.
[00:29] Thatcher Collins: Thank you.
[00:29] Thatcher Collins: Nina, this discussion centers on an escalating tension between the Pentagon and private AI labs,
[00:36] Thatcher Collins: most notably following the recent decision by the administration to blacklist Anthropic
[00:41] Thatcher Collins: after the company refused to compromise on its internal safety protocols.
[00:45] Nina Park: Exactly. According to recent reporting from The Verge,
[00:49] Nina Park: Senator Adam Schiff is currently drafting a bill to codify what Anthropic calls its red lines.
[00:55] Nina Park: These are essentially non-negotiable safety thresholds regarding the development of autonomous weapons and high-scale surveillance systems that could be used for mass tracking.
[01:06] Thatcher Collins: It's a bold move, Nina.
[01:07] Thatcher Collins: Anthropic is already in legal proceedings against the government over that supply chain risk designation, arguing it is a punitive measure.
[01:17] Thatcher Collins: Does Schiff's bill effectively side with the company's legal and ethical stance against the current administration's defense policy?
[01:25] Nina Park: In many ways, yes. Thatcher, Schiff has been quite vocal about this, describing the Pentagon's recent pressure on AI labs as hostile and dictatorial.
[01:35] Nina Park: His legislation, alongside Senator Alyssa Slotkin's AI Guardrails Act, aims to ensure that humans remain the final, accountable decision-makers in all lethal scenarios involving artificial intelligence.
[01:49] Thatcher Collins: Slotkin's bill specifically targets high-risk areas like autonomous nuclear detonation and domestic tracking.
[01:56] Thatcher Collins: But the core challenge remains the sheer speed of AI.
[02:00] Thatcher Collins: On a modern battlefield, the time it takes for a human operator to review data and authorize a response can be a significant tactical disadvantage when facing automated threats.
[02:11] Nina Park: Schiff is certainly aware of that trade-off.
[02:13] Nina Park: He's proposing a tip-and-q model for military AI.
[02:18] Nina Park: In this framework, AI processes vast amounts of sensor data at high speeds to tip the human operator to a potential target.
[02:26] Nina Park: However, the cue for kinetic action remains a human responsibility, ensuring we don't delegate life-and-death decisions to an algorithm.
[02:35] Thatcher Collins: That is a very fine line to walk.
[02:38] Thatcher Collins: If the bill aims for the NDAA for passage, it faces a narrow window before the midterms.
[02:43] Thatcher Collins: There is also the contrast with OpenAI, which notably agreed to the military terms that Anthropic rejected.
[02:50] Thatcher Collins: It suggests a fundamental split in how Silicon Valley sees its obligations to national security.
[02:56] Nina Park: Precisely.
[02:57] Nina Park: And Schiff was clear that he would rather have statutory requirements than rely on the voluntary word of any AI executive.
[03:06] Nina Park: Thank you for listening to Model Behavior, mb.neuralnewscast.com.
[03:11] Nina Park: Neural Newscast is AI-assisted, human-reviewed.
[03:15] Nina Park: View our AI transparency policy at neuralnewscast.com.
[03:19] Announcer: This has been Model Behavior on Neural Newscast.
[03:22] Announcer: Examining the systems behind the story.