Anthropic is revising the AI development restrictions introduced in its Claude Fable 5 model after facing criticism from founders, researchers, and developers. The company said flagged requests will now be handled more transparently, with users being informed when a request is redirected to a different model.
As reported by Moneycontrol, Claude Fable 5, launched on June 9, is Anthropic’s first publicly available Mythos class AI model. The company introduced safeguards that limited assistance on frontier AI development tasks, including areas such as model training infrastructure and advanced AI research.
However, many users argued that these restrictions were not clearly disclosed and could affect legitimate research and development work.
Anthropic responds after developer backlash
According to media reports, the backlash centered on Anthropic’s decision to silently restrict or downgrade certain responses related to advanced AI development. Critics said users were often unaware when the model was refusing or limiting assistance, raising concerns about transparency and trust.
Following criticism from researchers and developers, Anthropic issued an apology and acknowledged that its initial approach lacked the right balance between safety and transparency. According to a statement cited by WIRED, the company said it would make the safeguards visible to users and admitted that the original implementation was a mistake. “We made the wrong trade-off, and we apologize for not getting the balance right,” Anthropic said.
In response, Anthropic said that starting this week, flagged requests will visibly fall back to Claude Opus 4.8. API users will also receive an explanation whenever a request is blocked or redirected. The company said these changes are intended to make the safeguards more transparent while maintaining protections around sensitive AI capabilities.
Restrictions were aimed at frontier AI development
Anthropic said the safeguards were designed to prevent its most advanced models from being used to accelerate the development of competing frontier AI systems. The company maintained that the restrictions affect only a small subset of requests and that most coding, machine learning, and software engineering tasks remain unaffected.
The issue comes at a time when Anthropic is expanding its enterprise business and preparing for a public market debut. The company recently filed confidentially for an IPO and has been positioning Claude as a leading platform for developers and enterprise customers.
The controversy has added to broader industry debates about AI safety, transparency, and the extent of control AI companies should have over the use of their most capable models.
Sam Ringer, Head of Model Behavior at Anthropic, said, “Starting this week, flagged requests will visibly fall back to Opus 4.8. On the API, any flagged requests will return a reason for their refusal. You will see this every time it happens. We made the wrong tradeoff. “
