Anthropic has introduced Claude Sonnet 5, a more capable option for developers building autonomous "agent" systems that can plan multi-step workflows, use tools, and execute tasks with limited human input. Sonnet 5 is designed to operate more independently than prior Sonnet versions, including the ability to interact with tools such as web browsers and terminals to complete longer, more complex task chains, the company said in a blog post.

The model is rolling out across its product suite, where it will serve as the default for Free and Pro users. It will also be available to Max, Team and Enterprise customers, as well as through Claude Code and the Claude Platform API under the name claude-sonnet-5.

Safety and Security Features

On safety, Anthropic said internal testing indicated Sonnet 5 produces fewer undesirable behaviors than Sonnet 4.6 and shows improved resistance to prompt injection attacks and malicious instructions. The company also emphasized that Sonnet 5 was not explicitly trained for cybersecurity applications and performs significantly below its top-tier Opus models on evaluations involving potentially harmful cyber capabilities.

"We did not deliberately train Sonnet 5 on cybersecurity tasks. It can perform some routine, non-harmful cyber tasks, but on evaluations testing potentially dangerous cyber skills, such as developing software exploits, it shows substantially poorer performance than models such as Opus 4.8 and Mythos 5," the company said.

Anthropic said it has enabled cyber safeguards by default in Sonnet 5, using real-time detection and blocking systems similar to those deployed in its Opus models, while noting the guardrails are less restrictive than those applied to Fable 5.

Rate Limits and Availability

The company also raised rate limits across Chat, Cowork, Claude Code and the Claude Platform, citing higher token usage associated with increased "effort" settings in the new model. The broader availability aims to support developers integrating agentic AI capabilities into their applications.

The U.S. Department of Commerce has lifted export controls on Anthropic's Claude Fable 5 and Mythos 5 artificial intelligence models, allowing the company to begin restoring access globally. Commerce Secretary Howard Lutnick stated that officials worked closely with Anthropic over the past two weeks to analyze and approve Fable 5, ensuring alignment across the U.S. government and strengthening America's leadership in AI. The decision resolves a national security-driven standoff that had previously restricted access to these models for foreign nationals.

Access Restoration and Rollout Plan

Anthropic confirmed that Fable 5 will begin rolling out globally on Wednesday across the Claude platform, Claude.ai, and Claude Code. The company plans to restore access through Amazon.com, Inc. Web Services, Alphabet Inc.’s Google Cloud, and Microsoft Corp’s Foundry as quickly as possible. Additionally, Anthropic will continue expanding access through its Glasswing cybersecurity program, which provides vetted organizations with advanced AI models for defensive security testing.

Regulatory Background and Safeguards

The restrictions were initially imposed in mid-June under a government directive issued under national security authorities. Prior to the complete lift, the Commerce Department had permitted the release of the Claude Mythos 5 model to select U.S. organizations after receiving government approval on June 26. Earlier, Lutnick had stated in a letter to Anthropic co-founder Tom Brown that appropriate safeguards were in place to permit certain trusted partners to access the Claude Mythos 5 model, while approval for the consumer-facing Fable 5 remained withheld.

Sector-Wide Regulatory Landscape

The regulatory environment has tightened across the AI sector, with rival OpenAI recently unveiling its GPT 5.6 Sol, Terra, and Luna models. OpenAI limited the initial distribution of these models to trusted partners at the government's request, driven by GPT 5.6's capabilities, which were considered comparable to Anthropic's Mythos model. Both OpenAI and Anthropic have confidentially filed paperwork as they prepare for potential initial public offerings.

Government Statements and Market Reaction

David Sacks, co-chair of the President's Council of Advisors on Science and Technology, had previously indicated that Fable 5 could return to general release once Anthropic resolved the "safety issue." President Donald Trump had also commented on the shifting threat assessment, noting the administration's continued evaluation of the balance between the model's capabilities and national security risks. Ahead of the approval, prediction markets on the federally authorized platform Kalshi reflected strong market interest in the resolution of these restrictions.