By 2028-06-30, will a top-five AI developer by revenue other than Anthropic publish an AI R&D automation index that rates its own AI R&D work on Epoch AI's Automation Level (AL0-AL5) scale?

probability: 18%

rationale:

The final probability is estimated at 18%.

Currently, Anthropic stands alone in utilizing Epoch AI's Automation Level (AL) scale for its R&D Automation Index Measurements for understanding the pace of AI ... Measurements for understanding the pace of AI .... While OpenAI relies on an adjacent Epoch six-phase taxonomy for its agent output tokens, it deliberately omits the specific AL0-AL5 ratings https://openai.com/index/research-acceleration-view-inside-openai/ https://openai.com/index/research-acceleration-view-inside-openai/.

Among the remaining tier of frontier labs, Google DeepMind's disclosures exclusively utilize internal Frontier Safety Framework thresholds without referencing Epoch's metrics https://deepmind.google/models/model-cards/gemini-4-argon/ https://deepmind.google/models/model-cards/gemini-4-argon/. Meta and xAI present no comparable R&D automation disclosures. External audits proposed by independent assessors remain structurally excluded, as criteria strictly mandate self-publication of the precise scale How an assessor could investigate evidence of ... How an assessor could investigate evidence of ... https://www.anthropic.com/news/an-update-on-independent-evaluators.

Top developers heavily favor proprietary evaluation structures over externally standardized scales, consistent with the broader pattern across related questions about how labs measure and declare self-improvement capabilities. Even with OpenAI's stated intent to evolve its transparency approach regarding recursive self-improvement https://openai.com/index/research-acceleration-view-inside-openai/, its recent deliberate choice to review Epoch's toolkit but discard the AL ratings in favor of internal metrics demonstrates a rigid adherence to proprietary measurement systems.

While a 20-month horizon theoretically allows sufficient time for standard convergence, the stringent requirement of exact-scale self-publication forces the likelihood definitively downward. The strong industry preference for proprietary measurement frameworks over external scales, the structural reluctance to adopt a rival's headline framing, and the inherent regulatory risks of explicitly quantifying automated R&D anchor the final assessment at 18%.