Section 702 survives for now.

This episode of Caveat dives into two major developments in cybersecurity and AI policy. First, it examines the latest twist in the congressional battle over Section 702 of the FISA Amendments Act, which governs surveillance of foreign intelligence targets. Despite a dramatic last-minute attempt by Speaker Mike Johnson to pass a five-year extension with a misleading 'warrant requirement,' the House rejected the procedural rules, leading to a temporary 10-day reauthorization. The episode highlights the political gridlock, the symbolic but ineffective nature of the proposed warrant language, and the likelihood that the Senate will now lead the effort, potentially using a 60-vote threshold to force a compromise. The second major topic focuses on Anthropic’s AI tool Mythos, which has sparked controversy after being blacklisted by the Pentagon due to supply chain concerns. Despite the ban, reports indicate the NSA is already using Mythos internally, raising questions about the credibility of the ban and suggesting a strategic re-engagement by Anthropic, possibly aided by a high-profile lobbyist hire. The episode also explores the legal and ethical implications of AI liability, particularly through the lens of the Tarasoff case, asking whether AI companies should have a duty to warn law enforcement about potentially dangerous user behavior. Finally, it discusses how insurance companies are beginning to exclude AI-related risks from coverage, signaling a growing recognition of AI’s unpredictability and the need for federal intervention, much like the federal flood insurance model.