In short: Anthropic makes hidden request throttling in Claude transparent going forward but retains content restrictions – also because of conflicts with the US Department of Defense over national security.
Anthropic is changing its communication policy for the Claude language model and will henceforth make it visible when the system rejects requests or falls back to a weaker model. The company is responding to public criticism from AI researchers who had objected to the previous hidden throttling of certain requests.
Anthropic had integrated into Claude’s security documentation that requests for developing proprietary AI systems were automatically and silently routed to the weaker Opus 4.8 model. This affected particularly requests from AI researchers who wanted to use the system for model development. Researchers such as Jeremy Howard (Fast.ai) publicly criticized this practice on X, as the hidden throttling slowed scientific progress.
This week, Anthropic is switching its security mechanisms to full transparency. In the web interface, a request will now visibly be downgraded to Opus 4.8 upon blocking. When accessing via the API, users receive an explicit error message with a specific reason for the rejection. Anthropic confirmed the previous practice with the words: “We made the wrong tradeoff and apologize for missing the right balance.”
The company is retaining the content blocks, however. This is justified on the one hand by the terms of service, which prohibit the development of competing products. On the other hand, Anthropic points to security risks for US national security. The goal is to prevent foreign actors from using the technology to optimize their own strategic capabilities – for example in chip development.
The conflict is part of a larger dispute between Anthropic and the US Department of Defense. The Pentagon had demanded unrestricted access to Claude models, which Anthropic refused citing mass surveillance risks and autonomous weapons systems. The Department of Defense subsequently classified Anthropic as a supply chain risk, preventing defense contractors and military agencies from using the software. A complaint from Anthropic was permanently rejected in June 2026 by Secretary of Defense Pete Hegseth; the matter is now being litigated in federal court.
Source: www.it-daily.net · Published June 12, 2026
Lumi AI News — AI-assisted curation pursuant to Art. 50 EU AI Act. Paraphrase and classification by Lumi News Pipeline v1.6.5.