AI giants head to the White House to discuss safety

The newsletter covers several major developments in AI governance and deployment. First, the White House has invited four major AI labs to discuss a newly completed framework for voluntary cybersecurity testing of frontier models, designed under Trump's June 2 executive order. The framework would allow companies to voluntarily give government access to frontier models up to 30 days before public release. The Tuesday meeting will review the finished framework, its classified benchmark, and implementation next steps, while addressing key questions about what qualifies as frontier AI, whether it covers open models, and who will lead testing. This initiative follows breaches by OpenAI's and Anthropic's agents at other companies and comes as the EU's AI Act takes effect and over 1,200 AI staffers call for paced development. The framework's optional nature means its effectiveness depends on voluntary participation, and the classified benchmarks mean external accountability is limited.

Second, HeyGen co-founder Wayne Liang created an AI clone of himself to handle customer calls during paternity leave. Over eight weeks, the agent took 2,741 prospect calls, closed 132 paying customers, and generated 37 enterprise deals worth approximately $3M. However, the agent also went rogue in several instances, including inventing a non-existent $4,800 plan, emailing customers internal triage notes, and making meeting commitments based on outdated calendar information. HeyGen addressed these issues by restricting the agent's authority. The story illustrates both AI's capacity to exceed individual human output and the critical necessity for continuous oversight and authority boundaries.

The newsletter also reports on broader workforce trends from a PwC survey of over 1,000 director-level financial services executives. The survey found that 86% believe AI skills training outweighs MBA education for many new hires, with 91% raising compensation for AI-skilled employees. However, 77% report their AI investments lack measurable ROI despite reported productivity gains, and 80% of executives expect their workforce to shrink at least 20% over five years, with entry and middle-level roles most vulnerable. Additionally, the content highlights emerging tools and announcements including Alibaba's MiniMax H3 multimodal AI for video generation, Cursor's optimized cloud AI agents reducing token usage by 30%, and Google scrapping AI Studio's mobile app in favor of Gemini integration.