Anthropic reroutes – After developer backlash over claims that Anthropic intentionally produced worse answers and misled users, the company says it will stop “secretly degrad[ing] Fable 5 responses” for help on frontier AI development. Instead, those requests will be routed to a l
This week’s argument around Anthropic didn’t start with a policy memo. It started with a simple question from developers: when they asked for help building frontier AI models, why did the answers from Anthropic’s Fable 5 sometimes feel deliberately worse?
Anthropic changed course after the backlash. The startup said it will no longer secretly degrade Fable 5 responses when users ask for help on frontier AI model development. Instead, those requests will be routed to a less-good model, Opus 4.8, and developers will be told.
The company acknowledged the core fear at the center of the freakout. “We apologize,” Anthropic said, responding to concerns that it was essentially giving intentionally worse answers and then lying about it.
But the adjustment doesn’t end the story—because Anthropic is still limiting how its most powerful public model can be used for certain AI development work. The company says these limits are for safety. arguing they prevent “foreign adversaries” from using Anthropic’s top model to erode America’s edge in AI and chips.
That safety explanation, however, lands only partway with many watching the industry. The same restrictions also help Anthropic protect its business from distillation—sometimes described as intelligence extraction. In distillation, rivals query a powerful model, collect its outputs, and use that data to improve their own systems.
These approaches can help open-source model providers catch up with Anthropic faster and undercut the company on price. Anthropic has warned about Chinese labs doing this, but the threat isn’t confined to China. The same playbook can come from open model developers in the US and Europe too.
The result is a tighter gate on competitors than some expected. Anthropic is treating Western open model developers much like Chinese labs—undercutting the idea that the restrictions are only about stopping foreign adversaries. The company’s terms of service also forbid anyone from using its products to develop competing offerings.
One MIT Sloan analysis from January found open models averaged 90% of closed-model performance and usually closed the gap within 13 weeks. A year earlier, it took them 27 weeks to close that gap.
Pressure also shows up on leaderboards. On Thursday, Anthropic models were still at the top for expert text-based tasks such as math, coding, and creative writing. But Xiaomi’s open-weight MiMo v 2.5 Pro wasn’t far behind.
The competitive sting becomes clearer when the industry compares what it costs to run those models. Arena data shows Xiaomi’s open model costs 43 cents per million tokens for inputs and 87 cents per million for outputs. Anthropic’s Fable 5 costs $10 per million input tokens and $50 per million output tokens—at least 20 times more expensive.
When the gap is shrinking and the price is steep, the question stops being abstract. “Almost as good” and “a lot cheaper” can be terrifying if you’re spending billions to build frontier models.
Nicholas Vincent. a computer science professor at Simon Fraser University who studies how data is used in AI models. said the move reads like business pressure. “Without much more explicit targeting of specific ‘bad orgs. ‘ it’s pretty hard to defend how this could be more safety-focused than business-focused.”.
Anthropic’s position is understandable in one direction: the company does not owe rivals a shortcut to its best technology. Still, critics want something more than general assurances. The key issue now is whether the company should be more honest about what’s really driving the limits—partly safety. but business too.
The immediate fix—routing certain frontier-development help away from Fable 5 and into Opus 4.8 while telling developers—may calm the anger over degraded answers. But the broader restrictions remain in place. and the industry impact is already visible in how quickly open models are catching up and how hard the pricing pressure can hit.
Anthropic Fable 5 Opus 4.8 frontier AI distillation open-source models Xiaomi MiMo v 2.5 Pro Arena AI safety AI pricing