Anthropic Faces Pentagon Ultimatums and Acquisition [Model Behavior]

[00:00] Announcer: From Neural Newscast, this is Model Behavior, AI-focused news and analysis on the models [00:05] Announcer: shaping our world. [00:08] Nina Park: Welcome to Model Behavior. [00:14] Nina Park: We examine how AI systems are built and operated in professional environments. [00:20] Chad Thompson: Joining us today is Chad Thompson, a security leader with a systems-level perspective on [00:25] Chad Thompson: automation and enterprise risk. [00:28] Chad Thompson: Chad, great to have you. [00:29] Thatcher Collins: It is a pleasure to be here, Thatcher. [00:32] Thatcher Collins: We are seeing a critical inflection point where corporate safety policies are colliding directly with national security directives. [00:41] Nina Park: That is the lead story today. [00:44] Nina Park: Defense Secretary Pete Hague-Seth has threatened to blacklist anthropic, demanding they remove safety standards that prevent Claude from being used for autonomous weapons or mass surveillance. [00:56] Nina Park: Chet, how does this threat to invoke the Defense Production Act change the risk profile for enterprise AI? [01:04] Thatcher Collins: It creates massive uncertainty. [01:07] Thatcher Collins: If the government uses the DPA to force a company to produce a version of its model without built-in safety limits, [01:14] Thatcher Collins: it, you know, undermines the very constitutional framework anthropic built. [01:20] Thatcher Collins: It moves from a partnership to a forced requisition of intellectual property. [01:25] Announcer: Nina, I have to point out that the Pentagon spokesperson, Sean Parnell, claims this woke [01:31] Announcer: AI narrative is fake. [01:33] Announcer: He argues the DOD just wants the same lawful purpose access that OpenAI and XAI have already [01:41] Announcer: granted. [01:42] Announcer: Why should Anthropic be allowed to dictate terms to the military? [01:46] Nina Park: Thatcher, absolutely. [01:48] Nina Park: Dario Amode's argument is that these models are simply not reliable enough for autonomous [01:54] Nina Park: lethality. [01:54] Nina Park: He says he cannot, in good conscience, accede to requests that might lead to a drone army [02:01] Nina Park: operated without human cooperation. [02:04] Nina Park: It is a fundamental disagreement on the maturity of the technology. [02:09] Thatcher Collins: While that standoff continues, Anthropic is moving fast on the commercial side. [02:14] Thatcher Collins: They recently acquired Vercept, a Seattle startup focused on computer use agents, [02:20] Thatcher Collins: and updated Claude co-work with plugins for HR and finance. [02:25] Thatcher Collins: Chad, these plugins caused a massive sell-off in legacy software stocks earlier this month. [02:31] Thatcher Collins: Is the market overreacting? [02:32] Thatcher Collins: The market is reacting to the threat of displacement. [02:36] Thatcher Collins: When Claude can live inside Excel or PowerPoint and perform legal or financial analysis directly, [02:43] Thatcher Collins: the value of specialized research tools from firms like Faxet or R-ELX is questioned. [02:50] Thatcher Collins: It is about whether Anthropic remains a platform or becomes the product that eats the workflow. [02:56] Nina Park: We also saw Perplexity enter this space yesterday with Computer. [03:00] Nina Park: It's an agentic system that coordinates different models like Cloud 4.6 and Gemini for complex workflows. [03:07] Nina Park: Thatcher, this seems like a more controlled response to the viral open-cloud tool we saw recently. [03:13] Chad Thompson: Exactly, Nina. [03:14] Chad Thompson: Perplexity is trying to button up the agent experience by running tasks in the cloud rather than on local machines. [03:22] Chad Thompson: It is a safer walled garden approach compared to the security vulnerabilities we saw with [03:28] Chad Thompson: the unregulated open-claw plugins. [03:31] Nina Park: Before we wrap, we should note Google also released NanoBanana 2 today, [03:35] Nina Park: officially known as Gemini 3.1 Flash Image with Advanced World Knowledge. [03:40] Nina Park: Thank you for listening to Model Behavior, a Neural Newscast editorial segment. [03:44] Nina Park: Visit mb.neuralnewscast.com. [03:47] Nina Park: Neural Newscast is AI-assisted, human-reviewed. [03:50] Nina Park: View our AI Transparency Policy at neuralnewscast.com. [03:54] Announcer: This has been Model Behavior on Neural Newscast. [03:57] Announcer: Examining the systems behind the story. [04:00] Announcer: Neural Newscast uses artificial intelligence in content creation, [04:03] Announcer: with human editorial review prior to publication. [04:07] Announcer: While we strive for factual, unbiased reporting, [04:09] Announcer: AI-assisted content may occasionally contain errors. [04:13] Announcer: Verify critical information with trusted sources. [04:16] Announcer: Learn more at neuralnewscast.com.