Anthropic's "Fable 5": Powerful AI with New Safeguards

Anthropic has launched "Fable 5," a new AI system it previously claimed was too powerful to release. This system includes safeguards to address potential threats identified in its original model, "Mythos." What's interesting is that Anthropic also plans to release "Mythos 5" to security experts and other vetted users. This version has some safeguards lifted. Anthropic had previously released an earlier version of Mythos under similar terms, as part of "Project Glasswing," which focuses on protecting companies from powerful AI models, especially regarding cybersecurity risks. Anthropic acknowledges the risks of releasing such a capable model, stating that without safeguards, its cybersecurity abilities could be misused. For dangerous queries, the system will sometimes defer to a less capable model, "Opus 4.8." This happens for requests related to cybersecurity, biology, chemistry, or distillation. The company emphasizes that these new models can "do profound good," citing their help in cybersecurity and accelerating life sciences research. This launch follows criticism of Anthropic's "Opus 4.7" for being less capable than its predecessor. The bottom line: This release highlights the ongoing tension between AI capabilities and the need for responsible deployment.