What smart people are saying about the 2 most controversial parts of Anthropic's new models

Anthropic's recent launch of its AI models, Claude Fable 5 and Mythos 5, has sparked significant debate among experts in the tech community. The company introduced these new models alongside surprising safeguards designed to mitigate risks associated with advanced AI. Notably, the models may intentionally reduce their assistance if they suspect users are engaging in frontier AI research, and some requests might be redirected to less capable versions. In a system card, Anthropic explained that these precautions aim to prevent powerful AI technologies from aiding users in developing competing models or enhancing potentially dangerous capabilities. However, critics have voiced concerns that these measures could hinder researchers, concentrate power within a few leading AI labs, and compromise response quality without users being aware. David Kasten, head of policy at Palisade Research, expressed cautious optimism about Anthropic's intentions, stating that the company appears genuinely focused on addressing the risks associated with Mythos. "I believe they are trying hard to mitigate what they identify as risky features," Kasten remarked. He acknowledged that releasing such a model carries inherent risks, likening it to a constant game of cat-and-mouse between attackers and defenders in the AI space. On the other hand, Davi Ottenheimer, a notable digital safety expert, challenged the initial warnings about Mythos, questioning whether it was as dangerous as Anthropic had previously claimed. He pointed out that the model, which was initially withheld from public access due to safety concerns, is now being sold unchanged. Roman Stanek highlighted that the real issue in cybersecurity isn't the capabilities of AI itself, but rather the longstanding vulnerabilities that have been known for years yet remain unaddressed. He contended that financial incentives have played a significant role in the lack of action on these vulnerabilities. Elie Bakouch criticized Anthropic's decision to intentionally limit Mythos's performance on certain AI development tasks, calling it detrimental to the research community. Meanwhile, Jeremy Howard warned that these safeguards could exacerbate concentration within the AI industry, as they allow leading labs to leverage their top models for frontier research. Patrick Moorhead shared his underwhelming experience with Fable 5, noting that the model declined to assist with certain tasks it deemed too risky. Gergely Orosz added that the safeguards may inadvertently impact those not involved in developing competing models, suggesting that Anthropic's approach could limit opportunities for many. Deedy Das, however, praised the capabilities of Claude Fable 5, describing it as a groundbreaking model that raises concerns about the future of software engineering. He emphasized its potential in complex tasks, including significant codebase migrations and advanced graphics generation, while also noting its competitive pricing compared to other leading models.