thezvi.substack.com/p/anthropic-does-not-understand-why
2 corrections found
Eight states suing OpenAI are asking to enjoin against any further testing of advanced AI models until there are adequate safeguards, and also against creation of models capable of recursive self-improvement ‘until they can ensure that such models will be aligned with human values.’
This overstates what the multistate OpenAI lawsuits seek. The blanket injunction against further advanced-AI testing and recursive self-improvement appears in Arkansas’s case, not in all eight states’ suits.
Full reasoning
The article attributes one specific injunction request to all eight state lawsuits against OpenAI. But the states filed separate cases, and the record shows that this particular remedy was sought in Arkansas’s case, not uniformly across all eight.
- The Arkansas attorney general’s official release says Arkansas’s lawsuit seeks an “Injunction from further advanced AI testing and creating models capable of recursive self-improvement until adequate safeguards are in place.”
- By contrast, Nebraska’s official complaint asks the court to enjoin OpenAI from violating Nebraska law and from the deceptive and unfair practices alleged in that case. Its prayer for relief does not ask for a blanket halt on all advanced-AI testing or a ban on creating recursively self-improving models.
- Reporting on the litigation likewise describes the actions as separate state lawsuits, not a single uniform request for relief across eight states.
So it is inaccurate to say that “Eight states” are all asking for that Arkansas-specific injunction. A more accurate formulation would be that at least one of the state lawsuits—Arkansas’s—seeks that relief.
3 sources
- Attorney General Griffin Sues OpenAI for Misleading Arkansans over Safety of ChatGPT - Arkansas Attorney General
Arkansas’s lawsuit seeks relief through: ... “Injunction from further advanced AI testing and creating models capable of recursive self-improvement until adequate safeguards are in place.”
- State of Nebraska ex rel. Hilgers v. OpenAI — Complaint
Prayer for Relief: the complaint asks the court to “Enjoin Defendants from continuing to violate Nebraska’s Age-Appropriate Online Design Code Act and engaging in the deceptive, unfair, and unconscionable acts and practices described herein,” along with penalties and restitution; it does not request a blanket halt on advanced-AI testing or recursively self-improving models.
- Just the Docs: Eight States Sue OpenAI, Alleging It Misled the Public About ChatGPT’s Safety
The attorneys general of Arizona, Arkansas, Idaho, Iowa, Kansas, Montana, Nebraska, and West Virginia filed coordinated lawsuits against OpenAI in their own state courts on Oct. 14.
Switched to Responsible Scaling Policy V3, which abandons its unilateral commitments to slow or stop development if Anthropic is unable to provide sufficient risk assurance.
Anthropic’s RSP v3 did not eliminate all unilateral pause/slowdown commitments. The policy still says Anthropic will delay AI development and deployment in certain scenarios until it can make a strong safety case or match competitors’ safety posture.
Full reasoning
This claim overstates what changed in Responsible Scaling Policy (RSP) v3.
Anthropic’s own RSP v3 still contains unilateral commitments to delay development and deployment in some circumstances:
- In Appendix A, under the scenario where Anthropic is in the lead, the policy says: “We will require a strong argument that catastrophic risk is contained ... We will delay AI development and deployment as needed to achieve this” unless Anthropic no longer believes it has a significant lead.
- In another scenario, where competitors have strong safety measures, the policy says Anthropic will match or exceed that risk-reduction posture and “Until we are able to do so, we will delay AI development and deployment as needed to achieve this.”
Anthropic’s official RSP page also says that when it clarified the text in v3.1, it was making explicit something that was already true in v3: “even if not required by the RSP, we remain free to take measures such as pausing the development of our AI systems ... This was true of RSP v3.”
That means it is inaccurate to say RSP v3 “abandons its unilateral commitments” to slow or stop development. The policy may have changed those commitments’ structure or scope, but it did not remove them altogether.
2 sources
- Anthropic’s Responsible Scaling Policy
Anthropic’s official RSP page says that in the April 2, 2026 v3.1 clarification, “even if not required by the RSP, we remain free to take measures such as pausing the development of our AI systems ... This was true of RSP v3.”
- Anthropic’s Responsible Scaling Policy (version 3.0)
Appendix A states: “We will require a strong argument that catastrophic risk is contained ... We will delay AI development and deployment as needed to achieve this” and, in a separate scenario, “Until we are able to do so, we will delay AI development and deployment as needed to achieve this.”