Blab Tech
Microsoft Criticizes Anthropics Fable Model Restrictions Amid Export Control Shake-Up
Microsoft’s chief executive, Satya Nadella, told engineers on Wednesday that Anthropic’s limits on requests to its high‑end Fable model are ill‑founded. In a briefing recorded for CNBC, Nadella remarked, “If you use Fable, when it refuses for any random thing, it just is like, when was the last time you had a creation tool that was so editorially controlled?” He added that the restrictions “don’t make sense.” Microsoft declined to comment on the statement, and an Anthropic spokesperson did not respond to a request for comment.
The remarks came as Microsoft’s internal AI team was hard‑at‑work on the company’s Copilot product. Nadella’s comments were part of a broader discussion on how the firm is positioning its own models against external offerings, especially after the recent regulatory turbulence surrounding Anthropic’s flagship Claude line.
In early June, Anthropic announced Fable 5, a version of its Claude model engineered to cut false positives for blocked requests. Three days later, a U.S. government export‑control directive ordered the suspension of all access to Fable 5 and its private counterpart, Mythos 5, for any foreign national. The directive, issued on June 12, followed a revelation that a jailbreak method could bypass the model’s safeguards.
On July 1, Anthropic lifted the ban and restored Fable 5 globally. The company said the new safeguards would flag a slightly higher fraction of harmless requests than the previous version. A day later, on July 2, it released detailed documentation on its cybersecurity measures and a jailbreak framework.
Nadella’s comments arrive at a time when Microsoft is emphasizing cost‑efficient, internally developed models that can be tailored to corporate data. The company has invested $5 billion in Anthropic, while Anthropic has pledged $30 billion to Microsoft’s Azure cloud. Microsoft has also launched Copilot Cowork, a business‑productivity assistant that uses Anthropic’s models, and made the service generally available to Microsoft 365 customers.
Microsoft’s Foundry service, which offers developers access to more than 11,000 models—including those from Anthropic and OpenAI—provides a broader ecosystem for enterprises. Nadella has argued that relying on a small number of large‑cap companies for token‑based AI services is economically unsustainable.
The tension between Microsoft and Anthropic is not limited to model restrictions. In March, Microsoft announced that former Snap executive Jacob Andreou would lead Copilot across both consumer and corporate categories. Nadella described the move as something the company should have done “maybe day one.” The unification of Copilot products is part of Microsoft’s strategy to deliver a consistent AI experience across its product lines.
The broader AI landscape is also shifting. On Thursday, Chinese startup Moonshot AI unveiled an open‑source model that it claims outperforms recent releases from Anthropic and OpenAI. The announcement underscores the growing competition among AI developers and the increasing importance of open‑source models.
Several questions remain unresolved. It is unclear how Anthropic’s new safeguards will affect the model’s performance for legitimate use cases, and whether the export‑control directive will be lifted again. Microsoft’s stance on Anthropic’s restrictions may influence future partnership dynamics and the availability of AI tools for its customers.
In summary, Microsoft’s criticism of Anthropic’s Fable model limits reflects broader concerns about model control, cost efficiency, and regulatory compliance in the AI industry. The partnership between the two companies remains significant, but the recent events underscore the challenges of balancing innovation with safety and policy requirements.
The remarks came as Microsoft’s internal AI team was hard‑at‑work on the company’s Copilot product. Nadella’s comments were part of a broader discussion on how the firm is positioning its own models against external offerings, especially after the recent regulatory turbulence surrounding Anthropic’s flagship Claude line.
In early June, Anthropic announced Fable 5, a version of its Claude model engineered to cut false positives for blocked requests. Three days later, a U.S. government export‑control directive ordered the suspension of all access to Fable 5 and its private counterpart, Mythos 5, for any foreign national. The directive, issued on June 12, followed a revelation that a jailbreak method could bypass the model’s safeguards.
On July 1, Anthropic lifted the ban and restored Fable 5 globally. The company said the new safeguards would flag a slightly higher fraction of harmless requests than the previous version. A day later, on July 2, it released detailed documentation on its cybersecurity measures and a jailbreak framework.
Nadella’s comments arrive at a time when Microsoft is emphasizing cost‑efficient, internally developed models that can be tailored to corporate data. The company has invested $5 billion in Anthropic, while Anthropic has pledged $30 billion to Microsoft’s Azure cloud. Microsoft has also launched Copilot Cowork, a business‑productivity assistant that uses Anthropic’s models, and made the service generally available to Microsoft 365 customers.
Microsoft’s Foundry service, which offers developers access to more than 11,000 models—including those from Anthropic and OpenAI—provides a broader ecosystem for enterprises. Nadella has argued that relying on a small number of large‑cap companies for token‑based AI services is economically unsustainable.
The tension between Microsoft and Anthropic is not limited to model restrictions. In March, Microsoft announced that former Snap executive Jacob Andreou would lead Copilot across both consumer and corporate categories. Nadella described the move as something the company should have done “maybe day one.” The unification of Copilot products is part of Microsoft’s strategy to deliver a consistent AI experience across its product lines.
The broader AI landscape is also shifting. On Thursday, Chinese startup Moonshot AI unveiled an open‑source model that it claims outperforms recent releases from Anthropic and OpenAI. The announcement underscores the growing competition among AI developers and the increasing importance of open‑source models.
Several questions remain unresolved. It is unclear how Anthropic’s new safeguards will affect the model’s performance for legitimate use cases, and whether the export‑control directive will be lifted again. Microsoft’s stance on Anthropic’s restrictions may influence future partnership dynamics and the availability of AI tools for its customers.
In summary, Microsoft’s criticism of Anthropic’s Fable model limits reflects broader concerns about model control, cost efficiency, and regulatory compliance in the AI industry. The partnership between the two companies remains significant, but the recent events underscore the challenges of balancing innovation with safety and policy requirements.
