Anthropic introduced Claude Mythos Preview, a specialized model designed to target high-stakes cybersecurity and reasoning tasks. Access to the model is strictly limited to approximately 40 critical-infrastructure organizations to assist in locating vulnerabilities.