“The company ripped off a lot of writer’s work (per the $1.5B settlement), and recently walked back its core safety pledge.”
Our reading:
Anthropic recently walked back its core safety pledge.
Assessment: Supported
What happened
Was it true for the period described?
Outcome: Supported
The evidence matches what they said.
Judged on a plain reading. They named no measure, so we used the reading a typical informed reader would take. It is spelled out under “How we judged it”.
Evidence up to . Checked .
Why
Anthropic revised its Responsible Scaling Policy on February 24, four days before the post. The preceding policy tied training and deployment to required safeguards. The revision separated company plans from broader industry safety recommendations and expressly made roadmap goals nonbinding. It retained narrower pause commitments linked to competitive circumstances and introduced reporting requirements. That supports the ordinary meaning of walking back a central safety pledge, without implying that Anthropic abandoned safety altogether.
Caveats
- Marcus did not identify the precise policy clause; the Responsible Scaling Policy is the ordinary contextual reading.
- The assessment concerns the strength of a public commitment, not whether the revision increased actual harm or whether Anthropic violated the previous policy.
Evidence (3 items)
Supports the claim: Anthropic,
public goals
Why it matters: Anthropic explicitly made its roadmap goals nonbinding. The announcement explains that it was restructuring the earlier safeguards framework and separating achievable company plans from more ambitious industry recommendations, supporting a material weakening of the earlier pledge.
- Publisher
- Anthropic
- Kind of source
- Company disclosure: the company describing itself, not an independent finding
- About events on
- Published
Context only: Anthropic,
We will not train models
Why it matters: The preceding policy imposed training restrictions when required security safeguards were unavailable. This establishes the earlier commitment for comparison; its effective date precedes the target period.
- Publisher
- Anthropic
- Kind of source
- Company disclosure: the company describing itself, not an independent finding
- About events on
- Published
Context only: Anthropic,
high-level and limited
Why it matters: The revised policy retained commitments to delay development under specified conditions involving Anthropic's lead or competitors' safety measures, and added reporting obligations. This limits the finding to a weakened central pledge; it does not support a claim that all safety obligations disappeared.
- Publisher
- Anthropic
- Kind of source
- Company disclosure: the company describing itself, not an independent finding
- About events on
- Published
How we judged it
- Kind of claim
- Observation
- Deadline
- None given. We don’t add one.
Where we had to interpret
- The company refers to Anthropic. This assessment covers the normalized safety-policy claim, not the separate allegation about writers.
- The plain reading is that Anthropic materially weakened a central commitment to require adequate safeguards before advancing potentially dangerous AI systems. It does not mean Anthropic abandoned every safety commitment.
- I interpret 'recently' as the weeks preceding the February 28, 2026 post and use February 1–28, 2026 as the target period.
- I identify the core pledge as the Responsible Scaling Policy's commitment linking model training and deployment to adequate safeguards.
Right if: Dated policy documents show that, during the target period, Anthropic removed, narrowed, or made nonbinding a central requirement to establish adequate safeguards before proceeding with potentially dangerous model development or deployment.
Wrong if: The policy record shows that the relevant central commitment remained substantively intact throughout the target period, or that the supposed revision strengthened or clarified it without weakening its requirements.
We wrote this test after the outcome was known.
We found 6 claims in this post. See them all.
Full record
Source and dates
Marcus on AI, The whole thing was a scam, paragraph 12.
- Said
- We read the source
- live page, full text, public
- Evidence up to
- Checked
We read this wording on 2026-10-28. The post may have been edited after 2026-02-28.
Where it was said (1)
Said once in the posts we read.
: Marcus on AI, The whole thing was a scam
The company ripped off a lot of writer’s work (per the $1.5B settlement), and recently walked back its core safety pledge.
- Where
- paragraph 12
- Published
- 2026-02-28
- We read it
- , live page
- Access
- full text, public
History
Versions
- Current: Supported. Evidence up to 2026-10-28.
What happened to this record
PublishedBy us
Published after both AI reviews and the final check passed.