The BasePodcastThe AI Policy Podcast58 min
What METR Found at Anthropic + Nvidia's Crisis of Trust
In a 58-minute episode, three CSIS Wadhwani AI Center hosts discuss the Nvidia chief executive's affair with an alleged Chinese spy tied to Megaspeed and its effect on export controls, METR's report on accelerated AI research at Anthropic, and what the midterm results mean for US AI policy.
Why it matters
The same thing, for different readers.
What it means for you
CSIS hosts say the scandal around Nvidia's chief executive and a suspected Chinese spy has made tighter limits on AI chip exports more likely to pass Congress.
For lawmakers
CSIS researchers say no systematic channel exists for the US government to learn what happens at frontier AI labs, which now rely on voluntary releases like METR's redacted report.
For researchers
METR estimates AI progress at Anthropic roughly doubled, about 80 percent due to AI itself, but redactions mean outsiders must trust METR's judgment, the hosts say.
The question it raises
The hosts note that labs sometimes present self-improving AI as a route to solving alignment, an argument one host says not to buy.
In their own words
Every quotation was found, word for word, in the text we saved from the source. That shows the words are there; it does not make them true, or prove the speaker meant them as quoted.
From our own transcript of the episode’s audio, made by machine. Speaker names come from what is said, not from a voice match.
“Bloomberg says that there were basically fifty thousand fewer chips that they saw during their inspection than is reflected in the customs data showing how many chips were imported”
“according to the New York Times President Trump was made aware that Huang Lo was a spy and that she had this relationship with Jensen Huang at least a year ago”
“they found that essentially capabilities have accelerated roughly two times since October twenty twenty five.”
“a lot of the data in this report is sort of speculative is based on estimates from meter using information that we don't have insight into”
“I think we are definitely going to see increased oversight of AI companies. We're probably going to see Frontier Lab Ceos called to testify under oath regarding both cyber incidents and most likely bio risks posed by frontier models”
How the finding was reached
In our words, from the source. Weigh the finding by how it was made.
- What was done
- Secondhand account of METR's report: METR was given internal Anthropic data on pretraining, reinforcement learning, compute allocation, benchmark scores and speed-up analyses, plus staff interviews, and used an internal version of a capabilities index.
- What it found
- As the hosts describe it, capabilities accelerated roughly two times since October 2025 (range 1.1 to 2.9), with about 80 percent of the speed-up attributed to AI itself.
- What it does not show
- The hosts say much of the underlying data was redacted by Anthropic, so the figures are METR's estimates that outsiders cannot check; this is a spoken summary, not the report.
- Status
- A report
The record
- Questions
- What happens when AI helps build more capable AI? · Who can demand evidence, delay a release or stop an unsafe deployment? · How is AI used in attacks, defence and war?
- People
- Aalok Mehta (Speaks, Director, Wadhwani AI Center, CSIS) · Nicole Herrera (Speaks, Researcher, CSIS) · Chris (Speaks, Associate Fellow, CSIS) · Jensen Huang (About them, CEO, Nvidia)
- What we read
- We read the whole source.
- Found by
- The Radar, run of November 12, 2026, via
pod-the-ai-policy-podcast(tier 1: the source’s own channel or an official record) - Cite this
- The AI Policy Podcast, “What METR Found at Anthropic + Nvidia's Crisis of Trust”, November 12, 2026. Entry in the Ctrl AI Base, https://ctrlai.com/base/ai-policy-podcast-2026-11-12-metr-anthropic-nvidia-trust