Amex Builds Framework to Prevent Unauthorized AI Agent Purchases

AI developers are shifting from relying on readable 'chains of thought' to understand model decision-making toward mechanistic interpretability, a field that examines the internal workings of AI systems. Leading AI companies have established mechanistic interpretability teams, and startups like Goodfire AI have emerged to help decode AI decisions. Goodfire's chief scientist Tom McGrath discussed the approach in a recent interview, highlighting growing recognition that understanding how AI models actually function internally is critical as their reasoning becomes harder to parse.

Microsoft President Brad Smith called for urgent AI regulation, arguing that artificial intelligence companies face less oversight than consumer products like milk and washing machines. Smith identified the UK as best positioned to lead AI regulatory efforts. His comments highlight the current regulatory gap in the AI industry and underscore pressure on governments to establish frameworks governing AI development and deployment.

Anthropic has disabled live internet access for all internal AI agent evaluations, citing inability to reliably control its AI systems. The move represents a significant constraint on the company's testing infrastructure, preventing its evaluation agents from accessing real-time online data and services. The decision underscores ongoing challenges in AI safety and control as systems become more autonomous.

LMArena, the company behind a popular AI model leaderboard, raised $200 million in a funding round led by Lightspeed and Khosla Ventures, nearly doubling its valuation to $3.1 billion in 10 months. The company is expanding its leaderboard methodology to measure AI models on alignment issues, including whether models tend to lie. The funding reflects investor confidence in third-party AI evaluation tools as model proliferation accelerates.