Anthropic introduced its Responsible Scaling Plan at the end of 2023, setting thresholds for AI system capabilities.

Judge as of 2024-04-12; 'end of last year' means late 2023. True: Anthropic's official publication shows it introduced its Responsible Scaling Policy, setting capability thresholds (AI Safety Levels), in Q4 2023 (October–December). Mostly true: it was introduced in 2023 and sets capability thresholds, but slightly earlier than 'end of year' (e.g. September 2023), or the name differs slightly (Policy vs Plan). Mixed: introduced in 2023 well before the second half of the year, or it does not clearly set capability thresholds. Mostly false: the policy exists but was introduced in a different year. False: no such policy was introduced by that date.

Set before the research began.

1. E1 Introducing Anthropic's Responsible Scaling Policy Anthropic · 19 September 2023 Own record

   “Today, we’re publishing our Responsible Scaling Policy (RSP) – a series of technical and organizational protocols that we’re adopting to help us manage the risks of developing increasingly capable AI systems.”

2. E2 Introducing Anthropic's Responsible Scaling Policy Anthropic · 19 September 2023 Own record

   “Our RSP defines a framework called AI Safety Levels (ASL) for addressing catastrophic risks, modeled loosely after the US government’s biosafety level (BSL) standards for handling of dangerous biological materials.”

Anthropic's official announcement, dated 2023-09-19, says the company is "today" publishing its Responsible Scaling Policy (E1). The announcement says the policy defines AI Safety Levels, a tiered framework tied to catastrophic risk (E2), which matches the claim that it sets capability thresholds. The policy came out in September 2023, not late 2023, and its name is "Policy" rather than "Plan". The criteria list both of these differences under mostly true.