Step 13: assistant

text

I've assessed the claim as **supported**: credence 0.8 that it's true, and 0.7 confidence that "supported" is the right status. The claim is that factors other than AI likely contributed to Anthropic's faster capability progress in late 2025. It rests entirely on one source, METR's report of 5 November 2026, and the evidence behind it was redacted by Anthropic.

**What METR says.** Its attribution section says the late-2025 acceleration "was likely influenced by non-AI factors". Its executive summary goes further: other causes "likely played a large role". METR measured a 1.9x acceleration in Anthropic's benchmark progress since October 2025, and guesses that about 80% of the October 2026 acceleration is due to AI.

**Why it is supported.**
- The claim only says non-AI factors contributed, not that they dominated, so it needs relatively little evidence.
- METR judges that AI's contribution grew over the period, so a non-zero non-AI share early on is the expected pattern.
- Among the candidate causes METR examined were hiring, data improvements, new algorithms and removing other bottlenecks.
- METR notes Anthropic's growth before late 2025 was unusually low, so some of the measured acceleration may be catching up.
- METR is confident faster growth in training compute was not a cause.

**Why it is not verified.**
- Anthropic redacted the details of these factors and METR's cumulative attribution estimates, so readers have only METR's word for evidence it saw and cannot publish.
- The claim assumes the late-2025 acceleration was real. One alternative modelling choice puts it as low as 1.1x, though most choices support it.
- Publishing the redacted factor analysis would settle it.

**Other changes:**
- **Structure:** I left the claim with no subclaims, since it is a single source's statement about redacted evidence.
- **Source record:** I added METR's authors and the date to the claim's one source. I also recorded that its passage appears word for word in the stored report, and that it reports results from evidence it has not published.
- **Importance:** I set importance to 0.35 and contestation to 0.3. It is a supporting point in the larger question of how much AI assistance drives Anthropic's acceleration, and few dispute it directly.

**Dependent claim.** The one claim that depends on this, that AI assistance caused most of Anthropic's acceleration since October 2025, is rated inconclusive and already weighs this point, so I didn't notify its steward. I logged the decision and answered both inbox messages. The pass cost about $0.27 of the $1.00 budget.