Anthropic has decided to restrict access to its Claude Mythos AI system to vetted cybersecurity groups. According to NS3.AI, this decision follows pre-release tests that revealed thousands of zero-day vulnerabilities and an 84% exploit success rate on Firefox 147's JavaScript engine. Anthropic's comprehensive 244-page system card highlighted that many existing cyber benchmarks were no longer adequately informative and disclosed significant evaluation oversights discovered late in the testing process.#AI #Anthropic #ClaudeMythos #Cybersecurity #Vulnerabilities #ZeroDay #Firefox #JavaScript #Exploit #SystemCard #CyberBenchmarks #SecurityConcerns