The Trump administration has lifted restrictions on Anthropic's Claude models after weeks of government scrutiny, with the company announcing on Tuesday night that Claude Fable 5 is now widely available globally. According to Anthropic, the Commerce Department had blocked foreign nationals from using both AI models on June 12, forcing the company to immediately take the products down for all users just days after they were unveiled. The restrictions were sparked by a report from cybersecurity researchers at Amazon, Anthropic's primary cloud computing provider, who 'had found a method of bypassing Fable 5's safeguards' that enabled them to discover and potentially exploit software vulnerabilities. As reported by Anthropic, officials have grown increasingly concerned since the company warned earlier this year that its Mythos model was adept at finding software flaws in a way that could be weaponized by malicious hackers and threaten critical computer networks around the world.
Anthropic has implemented an improved safety classifier that blocks the specific bypass technique identified by Amazon researchers in over 99% of cases. When the classifier triggers, requests are automatically routed to the less capable Claude Opus 4.8 model, with users receiving clear notifications. The company has adopted a deliberate 'safety margin' approach that flags not only clearly harmful requests but also many that are likely benign or ambiguous, particularly those involving cybersecurity analysis, vulnerability identification, or debugging tasks. Fable 5 will be available for users globally on multiple platforms including Claude Platform, Claude.ai, Claude Code, and Claude Cowork. For Pro, Max, Team, and select Enterprise teams, Fable 5 will be included for up to 50% of weekly usage limits through July 7, 2026, after which it will be available via usage credits. However, Fable 5 will not be immediately available on AWS, Google Cloud, and Microsoft Foundry, though Anthropic says that availability is coming 'as quickly as possible'. Mythos 5, which is exceptionally good at finding security vulnerabilities, is being restored to a select group of U.S.-based organizations approved by the federal government.
The model's return coincides with crypto experiencing its worst year for hacks, with losses exceeding $942 million across 121 security incidents in 2026. As reported by The Crypto Times, total value locked (TVL) across the DeFi ecosystem has fallen approximately 39% year-to-date, declining from roughly $115 billion in January to near $70 billion in recent weeks. The largest losses include Kelp DAO at $292 million from cross-chain message forgery and Drift Protocol at $285 million from a six-month social engineering operation that compromised administrative keys. Bridges have accounted for the largest share of losses, with North Korean groups linked to a significant portion of total losses. Despite the reduced TVL, the sector continues facing persistent threats from well-understood categories including misconfigurations, exposed keys, and flawed operational flows, rather than zero-day smart-contract exploits.
Under the enhanced safety framework, Fable 5 maintains remarkable speed and effectiveness while introducing operational overhead for security-related workflows. According to Anthropic's reports, the models can generate working proof-of-concept exploits for complex issues in under 31 minutes, and have found critical flaws across more than 1,000 open-source projects, including widely used components like the Linux kernel. However, the expanded safety margin means that prompts involving security reviews or vulnerability analysis are more likely to trigger classifiers, routing them to the less capable Opus 4.8 model or requiring rephrasing. Early developer feedback has been mixed, with some reporting continued strong performance on general architecture and optimization tasks, while others note increased guardrail encounters on security-adjacent queries compared with the brief initial launch window. The company's analysis shows that a single operator with this class of model could turn a month of software patches into working exploits in a single afternoon for a cost measured in few thousand dollars.
The lifting of restrictions on Anthropic's models comes as OpenAI also announced Friday that it is restricting the release of its new AI model, GPT-5.6 Sol, at the request of President Trump's administration. According to OpenAI, the new AI product would be accessible only to a select group of government-approved customers for a temporary period. This coordinated approach reflects the administration's broader concerns about AI capabilities that could be weaponized for malicious purposes. The restrictions highlight the growing tension between AI innovation and national security considerations, with both major AI companies facing similar scrutiny over their advanced language models' potential for cybersecurity applications.
