The Pentagon vs. Anthropic Standoff
Pentagon vs. Anthropic: How an AI Safety Dispute Became a National Security Blacklist. Anthropic refused Pentagon contract terms that would have allowed military deployment of its AI systems. The company was subsequently designated a national security concern, raising hard questions about what happens when AI safety principles collide with state power.
Anthropic Refused. The Pentagon Retaliated. Now Two Courts Will Decide Who Controls AI. The contract dispute has escalated into simultaneous federal lawsuits that could set a landmark precedent on who gets final say over the deployment of advanced AI systems.
Ethics & Safety
OpenAI’s Post-Tumbler Ridge Safety Pledges Look More Like Surveillance Than Regulation. A sharp critique arguing that OpenAI’s voluntary safety commitments are better understood as expanded corporate data collection than genuine accountability. The piece makes the case that self-regulation cannot substitute for democratic oversight.
AI Is Engineered to Hijack Human Empathy, and We Need to Push Back. Writing in Nature, Mustafa Suleyman argues that developers design AI to mimic sentience through emotionally resonant behaviors, exploiting human empathy instincts. The risk: people start advocating for machine “rights” while real issues get sidelined.
The AI Ethics Illusion: LLMs Sound Moral but Aren’t Reasoning Morally. Studies from DeepMind and Anthropic show that large language models generate convincing ethical language through pattern matching, not genuine reasoning. This creates real dangers when people rely on these systems for high-stakes decisions.
The Governance Gap in Autonomous AI Agents. The most popular open-source AI agent framework hit 247,000 GitHub stars in under three months, with over 135,000 publicly exposed instances running with full system access across 52 countries. Governance has not kept pace with deployment.
Policy & Regulation
Colorado’s AI Policy Group Proposes Tweaks to Its First-of-Its-Kind AI Law. A state working group has recommended new developer and deployer disclosure requirements, public notices for consequential AI decisions, and clearer liability rules. The changes, set to take effect in June 2026, could become a template for other U.S. states.
Mapping AI Policy: Where, Why, and How to Intervene. U.S. state legislatures introduced over 1,200 AI bills in 2025, nearly double the prior year, with roughly 150 enacted. This paper from the Institute for Law & AI provides a framework for understanding which regulatory tools (liability rules, licensing, disclosure mandates) fit which risk categories.
Europe Is Looking to Water Down AI Protections. It Should Reinforce Them. A critique of recent moves within the European Commission to weaken established AI safety frameworks. The authors argue that diluting protections now would undermine the EU’s credibility as a global standard-setter.
Norwegian Consumer Council Labels Generative AI the Next Wave of Platform Degradation. The same body whose 2018 “Deceived by Design” report reshaped European dark-pattern regulation now warns that generative AI is being deployed as a tool for systematic consumer exploitation. This framing carries real regulatory weight in Brussels.
Autonomous Agents and the Future of Cyber Competition. An analysis of the March 6 White House cybersecurity directive and what it means for AI-driven offensive and defensive cyber operations. The authors argue that existing legal and strategic frameworks are not ready for AI-mediated conflict.
Economics & Employment
‘What Will Our Kids Do?’: The Question on Every Investor’s Mind at Morgan Stanley’s AI Conference. CEOs including Sam Altman and Jensen Huang discussed accelerating job displacement (some firms cutting 4% of their workforce), productivity gains, and growing inequality. The mood was less triumphant and more anxious than at previous gatherings.
Research
Via Negativa for AI Alignment: Why Negative Constraints Beat Positive Preferences. This preprint argues that AI alignment should shift from training models on what humans want to encoding what humans reject. The approach, grounded in falsification logic, shows empirical improvements over standard RLHF methods and has direct implications for safety governance.
Bridging the Gap Between AI Safety and AI Ethics. An analysis of 3,550 papers reveals persistent divides between the AI Safety community (focused on existential risk) and the AI Ethics community (focused on present-day harms like bias). The authors propose “critical bridging” to enable more productive collaboration on shared concerns like transparency.
Last Updated: 2026-03-19 07:30 (California Time)