LWiAI Podcast #245 - TML-Interaction, Claude For Legal, Sam Altman on Stand

Last Week in AI released its 245th podcast episode, recorded on May 13, 2026 and hosted by Andrey Kurenkov and Jeremie Harris. The episode summarizes and discusses recent AI news across tools, business, policy, safety, synthetic media, and research.

According to the episode description, OpenAI released new voice intelligence API features including GPT Realtime 2, described as GPT-5-powered, plus realtime translation and Whisper transcription. The release emphasized the latency-reasoning tradeoff, larger context, and new guardrails amid fraud risks.

Thinking Machines previewed a low-latency, full-duplex conversational system with a two-model architecture and a custom inference stack. The description says it reported strong interactivity benchmark results but has no public access or third-party validation yet.

Anthropic pushed further into vertical products with Claude for Legal and deeper AWS availability, while the description notes ongoing ecosystem tension as platform model providers compete with application-layer companies.

In applications and business, the episode covers Sam Altman's testimony, Nvidia CEO Jensen Huang traveling to China with Trump after a last-minute invite, and AWS expanding its Anthropic partnership with the Claude Platform launch. It also covers a Chinese grey market selling Claude API access at 90 percent off through stolen credentials, model substitution, and harvesting users' prompts and outputs for resale as AI training data, operating through proxy networks described as transfer stations.

Isomorphic Labs, a DeepMind spinout, raised $2.1 billion to design drugs with AI. Projects and open source items include Petri, described as Anthropic handing its alignment toolbox to Meridian Labs with a 3.0 update, and an OpenAI item titled Daybreak.

Policy and safety topics include teaching Claude why, OpenAI's self-harm trusted contact feature in ChatGPT, an investigation of accidentally grading chain-of-thought during RL, criticism of Natural Language Autoencoders, and a review of the Risks from automated R&D section in the Anthropic Risk Report from February 2026. Synthetic media coverage notes George Clooney, Tom Hanks, and Meryl Streep backing a new Human Consent Standard for AI licensing. Research coverage includes METR saying Claude Mythos is testing the limits of AI evaluation.