Over 50 cybersecurity professionals have raised concerns regarding the recent U.S. government decision to ban access to Mythos 5 and Fable 5, the latest large language models (LLMs) developed by AI company Anthropic. In an open letter addressed to key officials, including Howard Lutnick, the U.S. Secretary of Commerce, and Sean Cairncross, the U.S. National Cyber Director, these industry experts have urged the government to reconsider its export control directives.
On June 12, Anthropic announced that the U.S. government had issued an order suspending access to the advanced LLMs for any foreign nationals. This directive forced Anthropic to temporarily halt access to both models for all customers to comply with government regulations. Fable 5, described as a general-access LLM, was built on the same sophisticated AI framework as Mythos 5, which is an upgrade from the previous Claude Mythos Preview. Anthropic highlighted that Fable 5 comes equipped with additional safeguards designed to mitigate misuse, particularly in sectors like cybersecurity, where malicious applications could lead to significant harm.
The U.S. government justified its decision with “national security concerns,” prompting speculation that it was fueled by research indicating methods to bypass Fable 5’s built-in guardrails. In response, Anthropic clarified that it reviewed demonstrations showcasing this technique being used to uncover a limited number of relatively minor vulnerabilities. The company noted that these vulnerabilities were simplistic and emphasized that other publicly available models could also identify them without the need for a bypass. Importantly, Anthropic rejected the claim of a “universal jailbreak” existing for Fable 5.
### Cybersecurity Community Responds
Within two days of the announcement, a collective of 54 Chief Information Security Officers (CISOs), along with various cybersecurity practitioners and vendors, formally expressed their dissent. Their open letter requested the lifting of the export control directives on Fable 5 and Mythos 5. Additionally, they called for a commitment from the U.S. government to manage AI risk assessments in a more open, scientific, and transparent manner in the future.
The signatories acknowledged that Anthropic’s models possess considerable capabilities for identifying vulnerabilities and weaponizing exploits; however, they argued that these models are not the only tools available for such purposes. The ability to detect insecure code is a foundational feature of any coding assistant, a capability that already exists in various other models, including OpenAI’s GPT-5.5, Anthropic’s Claude Opus, Sonnet, and Chinese models like Kimi 2.7.
Moreover, the professionals acknowledged Anthropic’s efforts to prevent the utilization of Fable for potentially harmful cyber operations, indicating a recognition of the company’s responsibility in addressing the concerns that influenced the U.S. government’s decision.
The letter conveyed a strong warning: “To withdraw the most capable tools from defenders without a justified reason, particularly when adversaries are advancing quickly, is inherently dangerous.” The cybersecurity experts expressed that such governmental action has not only generated market uncertainty but also risks diminishing America’s leadership in the AI sector—without any substantial rationale to support it.
Among the signatories were notable figures like Alex Stamos, chief product officer at Corridor and former chief security officer at Facebook and Yahoo; Joe Levy, CEO of Sophos; and Sandra McLeod, CISO at Zoom Communications. Their position has been echoed by other cybersecurity experts. For instance, William Wright, CEO of Closed Door Security, emphasized that while the U.S. government’s reactions indicate genuine concerns regarding the potential for jailbreaking these models, imposing a ban on access is not the correct solution.
Wright elaborated on the practical implications of the ban, stating, “Cutting off access to the model so abruptly will cause huge logistical problems, both within Anthropic and within any critical industry partners granted access.” He argued that such an abrupt change creates chaos, rather than fostering resilience in the cybersecurity landscape.
In his call for action, Wright urged the U.S. government to engage in transparent discussions with AI and cybersecurity experts, stressing the importance of clear guiding principles moving forward.
The situation illustrates a growing tension between national security concerns and the necessity for collaborative, transparent discourse in the rapidly evolving field of artificial intelligence. As cybersecurity continues to be pivotal in safeguarding digital infrastructures, decisions made in this arena will require careful balancing of innovation, compliance, and security oversight.
In conclusion, the public outcry from cybersecurity professionals underscores a call for not only lifting the ban on access to advanced AI models but also a re-evaluation of how AI-related policies are conceptualized and implemented in light of both security needs and industry advancements.