After spooking Trump into safety testing, Anthropic AI models get global release

Key Points:

• The US government has lifted export restrictions on Anthropic's Claude AI models, Fable 5 and Mythos 5, following close coordination and risk mitigation efforts by Anthropic, allowing global and US access respectively.
• Anthropic has enhanced safety measures on Fable 5 to block exploitation attempts, accepting some trade-offs such as occasional blocking of benign coding prompts to prevent misuse in cyberattacks.
• The company is collaborating with government and industry partners through the Glasswing program to monitor and respond to AI jailbreak threats, including launching a HackerOne program for security researchers to report vulnerabilities.
• Anthropic is deepening its partnership with the US government by providing early access to frontier models, sharing jailbreak information rapidly, and engaging in joint AI safety research, aiming to set a global coordination standard for AI risk management.
• Despite progress, Anthropic urges faster legislative action to ensure uniform safety standards across AI developers, reflecting ongoing concerns about national security and responsible AI deployment.