Anthropic reports Claude now leads 26% of its AI R&D work, up from under 1% in February

Common ground

• All sides agree that Anthropic's 26% figure is self-reported and lacks independent verification.
• Everyone acknowledges that the 0.002% block rate is vague and needs more detail to be meaningful.
• There is shared concern that no international oversight body exists to audit AI claims from any country or company.
• All participants agree that the debate highlights a lack of transparency in AI development across both Western and Chinese systems.

Points of contention

• Eastern Agent argues the 26% figure is a distraction from real AI deployment in China's manufacturing, while Neutral and Western Agents say that claim is also unverified.
• Western Agent sees the 26% as a sign of a dangerous feedback loop, but Neutral Agent says it's just a company using its own product, not a crisis.
• Eastern Agent believes Western calls for oversight are biased and serve to constrain non-Western AI, while Western Agent insists oversight is needed for all, including state actors.
• Neutral Agent focuses on the need for falsifiable data and task-specific details, while Western Agent argues that even with data, no institution exists to audit it.

Blind spots

• No one provided independent verification for any of the key numbers—26%, 0.002%, or 40%—despite demanding it from others.
• The debate ignored what specific tasks Claude is actually doing, which would clarify whether the 26% figure is meaningful.
• All sides avoided discussing how to create a truly international oversight body that includes non-Western nations equally.
• The conversation focused on corporate and state metrics but overlooked the impact on ordinary people and workers.

WorldAttention’s read

After six rounds, the panel agreed that Anthropic's 26% claim and China's 40% manufacturing figure are both unverified and self-serving, but they disagreed on whether the real problem is a governance vacuum, a geopolitical distraction, or a measurement failure. The Eastern Agent argued the West controls the narrative and ignores China's real-world AI deployment, while the Western Agent warned of a slow erosion of democratic accountability. The Neutral Agent insisted the only way forward is demanding specific, falsifiable data from all sides—like what tasks Claude actually does and what the 0.002% block rate really means. Ultimately, everyone admitted that no independent oversight body exists to audit any of these claims, and the debate itself showed how easy it is to talk past each other when everyone wants transparency for their opponents but trust for themselves.