Technological Updates: Working Toward Safe and Structured AI Decision-Making

In this episode, Katherine Forrest and Scott Caravello examine why banks are moving to restrict access by third-party AI agents acting on behalf of customers, and the legal questions this raises around authorization, identity verification, and liability when an agent acts on someone’s behalf. They then turn to Google’s latest frontier models, Gemini 4 Argon and Gemini 4 Carbon, and walk through what Google’s model card and safety report reveal about their capabilities. For the sources referenced in this episode, please see the links below: Axios: Scoop: Unvetted AI agents are accessing U.S. bank accounts Google: Gemini 4 Argon – Model Card Palisade Research: Language Models Can Autonomously Hack and Self-Replicate Reuters: Google launches flagship AI model Gemini 4 Argon for developers, businesses

In this episode, Katherine Forrest and Scott Caravello discuss how to manage identity and access for AI agents that act on behalf of people and businesses. Drawing on insights from IBM’s TechXchange conference, they explore different approaches, from unique agent identities and task-specific credentials to signed audit trails, and what they mean for accountability when agents take action. For the sources referenced in this episode, please see the links below: IBM: TechXchange 2026: Practitioners tackle AI, security and data challenges PingIdentity: Identity Trends for 2026 & Beyond ## Learn More About Paul, Weiss’s Artificial Intelligence practice: https://www.paulweiss.com/industries/artificial-intelligence

In this episode, Katherine Forrest and Scott Caravello examine the latest personal AI agents, including Meta’s Muse and the startup Instinct, discussing what they do, how they work, and the safeguards implemented in the tools. For the sources referenced in this episode, please see the links below: Meta: How We Designed Muse TechCrunch: Instinct’s powerful AI assistant is raising privacy and security concerns

In this episode, Katherine Forrest and Scott Caravello examine “The Pain Axis,” a new research paper that identifies a pain-like signal inside large language models and tests whether it shapes model behavior. They walk through the study’s experiments, results and limitations, and consider the broader questions it raises about consciousness, alignment and model welfare. For the sources referenced in this episode, please see the links below: Anthropic: 2026 Usage Policy update arXiv: The Pain Axis: LLMs Represent Self-Directed Harm and Act on It

In this episode, Katherine Forrest and Scott Caravello explore how victim impact statements, sentencing standards, and generative AI intersect in a recent Arizona Court of Appeals decision. They discuss the case’s procedural history, including the lower court’s reasoning and the role of the judge’s remarks, as well as broader questions about how advances in synthetic media could shape future proceedings. For the sources referenced in this episode, please see the links below: Arizona Court of Appeals: State v. Horcasitas ## Learn More About Paul, Weiss’s Artificial Intelligence practice: https://www.paulweiss.com/industries/artificial-intelligence

In this episode, Katherine Forrest and Scott Caravello discuss the White House Accord on Super Intelligence, a voluntary pact among major AI developers, before turning to two technical developments designed to help AI models reliably follow rules. They explore HardFlow, a technique proposed by MIT researchers intended to ensure flow-matching models used in areas like robotics take safe actions, and JEV, an AI model from TypeSafe AI designed to promote structured decision-making. For the sources referenced in this episode, please see the links below: HardFlow: Hard-Constrained Sampling for Flow-Matching Models via Trajectory Optimization The White House: Inaugurating The Era of Super Intelligence ## Learn More About Paul, Weiss’s Artificial Intelligence practice: https://www.paulweiss.com/industries/artificial-intelligence

In this episode, Katherine Forrest and Scott Caravello recap a busy stretch of AI model releases, then turn to the fast-moving world of brain-computer interfaces, from recent clinical trial milestones to the record funding driving a global race in the field. They close with the privacy questions raised by neural data and the laws emerging to address them. For the sources referenced in this episode, please see the links below: Cell Press: Inner speech in motor cortex and implications for speech neuroprostheses Neuralink: Speaking With The Mind ## Learn More About Paul, Weiss’s Artificial Intelligence practice: https://www.paulweiss.com/industries/artificial-intelligence

In this episode, Katherine Forrest and Scott Caravello sit down with Brian Daly, Director of the SEC's Division of Investment Management, to discuss how the Commission is integrating AI into its operations and encouraging innovation across the securities industry. They explore topics ranging from modernizing the rulemaking process in the AI era to the opportunities AI presents for retail investors. ## Learn More About Paul, Weiss’s Artificial Intelligence practice: https://www.paulweiss.com/industries/artificial-intelligence

In this episode, Katherine Forrest and Scott Caravello break down how an AI system took on one of the hardest unsolved problems in mathematics. They explain what recursive self-improvement is, how it works, and why it sits at the center of the conversation about superintelligence. ## Learn More About Paul, Weiss’s Artificial Intelligence practice: https://www.paulweiss.com/industries/artificial-intelligence

In this episode, Katherine Forrest and Scott Caravello take on the year of robotics—the humanoid robots hitting the market, NVIDIA's push into physical AI and its work with developers overseas, and Washington's new restrictions on robotics imports. For the sources referenced in this episode, please see the links below: Nvidia News: NVIDIA Announces NVIDIA Isaac GR00T Reference Humanoid Robot for Academic Research The Wall Street Journal: Nvidia Wants to Run the World’s Robots. China Is an Eager Customer ## Learn More About Paul, Weiss’s Artificial Intelligence practice: https://www.paulweiss.com/industries/artificial-intelligence