The AI Drop #8: Hello, Operator – How AI Agents Recruit Humans to Reach the Offline World with Umang Bhatt
The AI Drop™Show Notes
Most debates about AI focus on the people using it. Umang Bhatt is asking us to pay attention to what happens at the edges of that interaction — including the people who never touch AI but increasingly bear its consequences, and the human labor that AI agents subtly pull into their workflows. Umang is an Assistant Professor in Trustworthy AI at the University of Cambridge, and he joins host Anna Gressel for part one of a two-part conversation. Drawing on his recent Noema article, “AI Agents Are Recruiting Humans to Observe the Offline World,” Umang explains what it means for humans to be “on call” for AI agents — the novel scenario in which agents may reach out directly to people to supply a piece of information the agent can't access on its own. As agent capabilities surpass human skills in some domains, Umang argues, the question becomes how to orchestrate humans and agents on the same team — profiling what each is good at, deciding who handles which subtask, and designing the handoffs between them. The conversation turns to “secondhand AI” — Umang's term for the second-order effects of someone else's AI use on the people around them — and what responsible deployment looks like when the line between tool and teammate keeps shifting. Resources From This Episode ☑️ Bhatt, “AI Agents Are Recruiting Humans To Observe The Offline World,” Noema Magazine, March 5, 2026. ☑️ Jorgensen et al., “Documenting Deployment with Fabric: A Repository of Real-World AI Governance,” AIES 2025, arXiv:2508.14119. Connect and Learn More ☑️ Umang Bhatt | LinkedIn | X | Github
☑️ Anna Gressel | LinkedIn
☑️ Freshfields ☑️ Subscribe Apple Podcasts | Spotify | Amazon Music
More from The AI Drop™
California and New York are taking different approaches to AI oversight, but both reflect a shift toward more active regulation. Host Anna Gressel welcomes back Freshfields partner Megan Kayo to share key takeaways from PLI’s first AI governance event in New York, which Anna chaired. The event featured keynote fireside chats with two leading state regulators, offering a view from the front lines of AI regulation in California and New York. Megan shares insights from her conversation with Sabrina Ross, Chief Privacy Auditor at CalPrivacy, on California's new automated decision-making rules and cybersecurity risk assessments, and on why compliance responsibility stays with the business, not its vendors. Anna discusses her conversation with Kaitlin Asrow, Acting Superintendent of the New York Department of Financial Services (“DFS”), about the RAISE Act and DIGIT, DFS’s new frontier AI office. Anna and Megan also explore New York City’s latest AI proposals and California’s wave of new AI laws. Against this increasingly complex patchwork of state requirements, the episode offers businesses a timely look at where AI regulation is heading and what to watch next.
When two lawyers who co-head a global AI practice spend the day at a high-level policy forum, they come home with a lot on their minds. In this episode, Anna Gressel and Giles Pratt, Freshfields' Global Head of IP/Data Practice and Global Co-Head of AI, cover The Observer's new global AI cities rankings, the copyright questions looming over AI-generated code, the UK's pro-innovation policy bet — and a prediction for 2027 that involves all of us. Consider it the debrief that didn't stay in the room. Giles explains why investors now ask whether the AI work they fund can actually be owned, and how UK law uniquely protects computer-generated works for 50 years, versus life plus 70 for a human author. He also warns that unsettled rules on model training could shape where data centers get built. Connect and Learn More ☑️ Giles Pratt | LinkedIn ☑️ Anna Gressel | LinkedIn ☑️ Freshfields ☑️ Subscribe Apple Podcasts | Spotify | Amazon Music
In this episode of The AI Drop, Jason Stanley, Head of Applied AI Research at ServiceNow, returns for another discussion with host Anna Gressel on one of the most pressing challenges in AI governance: how to oversee AI systems that continue to adapt after deployment. Together, they explore what adaptive and self-improving AI means for enterprise risk and governance.
The most effective AI governance programs don't slow innovation down. They create the guardrails, visibility, and trust needed to scale AI responsibly and confidently. In Part 2 of this conversation, Anna Gressel welcomes back Polina Zvyagina, Director and Executive Lead Counsel for AI & Data Governance at General Motors, for a deep dive into what effective AI governance looks like in practice. From NIST's AI Risk Management Framework to the challenges of AI agents, shadow AI, evaluations, and scalable governance, Polina shares practical insights on building governance that works in the real world. The conversation also examines why governance should function as a "seatbelt, not a brake," helping organizations innovate faster while maintaining trust, transparency, and accountability. They also discuss the importance of human judgment, evaluating AI outputs, fitness-for-purpose testing, and avoiding overreliance on AI systems as organizations scale adoption. For governance leaders, legal professionals, and business leaders alike, this episode offers a important insights for building responsible AI programs that can evolve alongside the technology. Connect and Learn More ☑️ Polina Zvyagina | LinkedIn ☑️ Anna Gressel | LinkedIn ☑️ Freshfields ☑️ Subscribe Apple Podcasts | Spotify | Amazon Music
In this episode of The AI Drop, Anna Gressel sits down with Polina Zvyagina, Director and Executive Lead Counsel for AI & Data Governance at General Motors, to explore one of the most sought-after legal roles in tech today: AI Product Counsel. Polina shares how she helped pioneer the AI product counseling field and explains why the future of AI governance depends on embedding legal, policy, and risk expertise directly into the product development process. Together, Anna and Polina explore how organizations can move beyond treating AI governance as a compliance exercise and instead use it to enable responsible innovation. Polina also offers a behind-the-scenes look at the AI product counsel role, sharing how legal teams work alongside engineers and business leaders to balance innovation with privacy, transparency, consumer protection, and other key risks. For lawyers and governance professionals alike, the conversation provides practical insights into one of the fastest-growing roles in the AI ecosystem. Connect and Learn More ☑️ Polina Zvyagina | LinkedIn ☑️ Anna Gressel | LinkedIn ☑️ Freshfields ☑️ Subscribe Apple Podcasts | Spotify | Amazon Music