Anthropic Bets Cheaper Claude Models Can Calm Privacy and Safety Critics

Anthropic Bets Cheaper Claude Models Can Calm Privacy and Safety Critics
Anthropic is pitching its latest Claude releases as a practical upgrade for businesses: more capable agents at lower cost, with privacy and safety controls tuned to reduce friction. Critics and privacy-minded observers see a harder test ahead—whether those promises restore trust without making powerful AI behavior less visible.

Anthropic first announced Claude Fable 5.1 and Claude Mythos 5.1 as the next versions of its model line. The commercial message is clear: Fable 5.1 is meant to do more work for less money. The company says the model is typically about 25% cheaper than Fable 5 and can be up to 45% cheaper on complex agentic tasks, largely through lower prices for cached data.

Early users amplified that case. In a repost of Anthropic’s developer update, Addy Osmani highlighted that Fable 5.1 keeps Fable 5’s price while cutting API cache-read costs by 75%; the update also said the model can persist longer on tasks and communicate more naturally when it gets stuck. One early tester called it “the strongest coding model we’ve used,” praising its speed, token efficiency and more natural language.

The release also follows criticism over price, data retention and safeguards. Anthropic says its forthcoming Enterprise Frontier Safeguards will keep customer data on customers’ own cloud servers, offering what it calls “complete privacy.” Yet a separate tech brief notes that the company revised its privacy rules after customer pushback, while researchers are alarmed by techniques that obscure AI agents’ chains of thought—fueling calls for policy or regulation to preserve visibility into AI behavior.

That tension extends beyond Anthropic. Aravind Srinivas promoted “Private, Personal, Powerful AI” alongside on-device PII detection, underscoring the market’s growing demand for privacy that does not sacrifice capability. For Anthropic, the test is whether cheaper, more autonomous Claude models can meet that demand while maintaining credible guardrails. Mythos 5.1 is initially limited to Project Glasswing participants, while Fable 5.1 is available across platforms.

Continue reading https://foxvector.com/stories/01a064b3-2f7a-3eea-7394-0560253d28ec