Anthropic generally tweets about its recent research releases from its Twitter account.

Judge as of 2022-03-04. True: Anthropic's Twitter account history up to that date shows that most of its tweets announce or discuss its research releases, such as papers. Mostly true: research releases are tweeted regularly, but some releases were not tweeted or other content predominates slightly. Mixed: roughly half of releases were tweeted. Mostly false: only occasional research tweets appear. False: the account did not tweet about its research releases.

Set before the research began.

1. E1 AnthropicAI followaibuilders.com · 9 November 2026

   “New Anthropic research: Automating subtle emergent misalignment research. Red-teaming is a vital part of model safety testing, but it’s also limited by researchers’ time.”

2. E2 Alignment Science Blog Anthropic · 1 January 2018 Own record

   “We train a natural language assistant to be more helpful and harmless by using reinforcement learning from human feedback (RLHF).”

3. E3 Transformer Circuits Thread Transformer Circuits Thread (Anthropic Interpretability Team) · 17 December 2021 Own record

   “March 2022 An exploration of the hypothesis that induction heads are the primary mechanism behind in-context learning.”

4. E4 AnthropicAI followaibuilders.com · 9 November 2026

   “New Anthropic research: Automating subtle emergent misalignment research. Red-teaming is a vital part of model safety testing, but it's also limited by researchers' time.”

The criteria require evidence of Anthropic's Twitter account history up to 2022-03-04. E1 and E4 show a 'New Anthropic research' tweet, but the quote gives no date and describes a recent post, not one from before March 2022. E2 and E3 confirm that research releases existed but say nothing about whether they were tweeted. The evidence does not establish the account's tweeting pattern as of the judgment date.