Anthropic CEO Summoned to Pentagon Over Claude Use [Model Behavior]

[00:00] Nina Park: From Neural Newscast, this is Model Behavior, AI-focused news and analysis on the models shaping our world. [00:11] Chad Thompson: Welcome to Model Behavior. [00:13] Chad Thompson: Model Behavior examines how AI systems are built, deployed, and operated in real professional environments. [00:21] Chad Thompson: Joining us today is Chad Thompson, a director-level AI and security leader with a systems-level perspective on automation, enterprise risk, and operational resilience. [00:32] Chad Thompson: Chad, great to have you. [00:34] Nina Park: Today we are looking at a pivotal moment for Anthropic. [00:38] Nina Park: On February 21st, they launched Claude Code Security, a tool designed to scan code bases [00:45] Nina Park: for vulnerabilities by reasoning like a human researcher. [00:48] Nina Park: However, the news is overshadowed by Defense Secretary Pete Hegseth, [00:53] Nina Park: summoning CEO Dario Amode to the Pentagon tomorrow. [00:57] Nina Park: Nina, this appears to be an ultimatum regarding military use. [01:01] Thatcher Collins: From a systems perspective, this is a classic collision between Silicon Valley safety protocols and national security requirements. [01:12] Thatcher Collins: The Pentagon is threatening a supply chain risk designation because Anthropic is blocking surveillance and autonomous weapons applications. [01:23] Thatcher Collins: If they are banished, it creates a massive operational void in their existing $200 million contract. [01:33] Chad Thompson: That's exactly the risk, Chad. [01:35] Chad Thompson: While Anthropic faces government pressure, OpenAI is focused on enterprise expansion. [01:42] Chad Thompson: They just announced Frontier Alliances, a collaboration with McKinsey, BCG, and Accenture. [01:50] Chad Thompson: the goal is to help businesses build AI co-workers using their frontier platform. [01:56] Chad Thompson: Thatcher, they are also moving into physical infrastructure in a big way. [02:02] Nina Park: Correct, Nina. [02:03] Nina Park: Earlier this week, on February 19th, OpenAI partnered with India's Tata Group [02:10] Nina Park: to build the country's first large-scale AI data center. [02:15] Nina Park: It's part of the Global Stargate Initiative. [02:19] Nina Park: India is already their second largest market, with over 100 million users. [02:26] Nina Park: So they're securing the local compute capacity needed to support these new enterprise agents. [02:33] Chad Thompson: Absolutely. It's a massive logistical play. [02:37] Chad Thompson: At the same time, we're seeing a shift in how these models are powered and sold. [02:42] Chad Thompson: Google DeepMind had just put Lyria 3 into the Gemini app for music generation. [02:48] Chad Thompson: But Google's leadership is also warning the market. [02:51] Chad Thompson: Thatcher, Google Cloud VP Darren Morrie, had some pointed words for AI startups recently. [02:57] Nina Park: Mowry is warning that LLM wrappers and aggregators have their check engine light on. [03:04] Nina Park: He argues that simply slapping a UI on a third-party model is no longer enough. [03:10] Nina Park: We see this contrast in hardware, too, with the Toronto startup Talos. [03:15] Nina Park: They've built the HC1 chip, which hardwires model weights directly into silicon. [03:21] Nina Park: They claim it hits 17,000 tokens per second on Lama3.1. [03:26] Thatcher Collins: The Talos approach is fascinating because it targets the memory wall. [03:32] Thatcher Collins: By etching the model into the wiring, they eliminate the energy cost of moving data. [03:39] Thatcher Collins: This suggests a future where we have flexible clusters for training and hyper-efficient, hard-wired foundries or inference. [03:50] Chad Thompson: It suggests the industry is maturing into specialized tiers. [03:54] Chad Thompson: Whether it's Anthropics, researcher-level security tools, or OpenAI's consulting alliances, [04:00] Chad Thompson: the focus is shifting from what can the model do to how does it integrate into a resilient enterprise architecture or a national security framework. [04:10] Nina Park: For sure, Nina. The transition from general-purpose research to hardened, specialized deployment is the theme of this week's new cycle. [04:19] Nina Park: From data centers in Delhi to security standoffs at the Pentagon, [04:24] Nina Park: the operational stakes have never been higher. [04:27] Chad Thompson: Thank you for listening to Model Behavior. [04:30] Chad Thompson: Thank you for listening to Model Behavior, a Neural Newscast editorial segment. [04:36] Chad Thompson: Visit mb.neuralnewscast.com for more. [04:42] Chad Thompson: Neural Newscast is AI-assisted, human-reviewed. [04:47] Chad Thompson: View our AI Transparency Policy at NeuralNewscast.com. [04:52] Nina Park: This has been Model Behavior on Neural Newscast. [04:55] Nina Park: Examining the systems behind the story, Neural Newscast uses artificial intelligence in content [05:02] Nina Park: creation with human editorial review prior to publication. While we strive for factual, [05:07] Nina Park: unbiased reporting, AI-assisted content may occasionally contain errors. Verify critical [05:13] Nina Park: information with trusted sources. Learn more at neuralnewscast.com.