Last month, global banks, tech companies, and governments reacted urgently to the cybersecurity risks posed by Mythos, an AI model from Anthropic claimed to have identified thousands of unknown software vulnerabilities. Despite the heightened concern, cybersecurity experts noted that these vulnerabilities can also be detected using existing AI models from firms like Anthropic and OpenAI. In response to these developments, OpenAI has introduced a limited-release cybersecurity model to tackle the challenges presented by Mythos. Anthropic has restricted access to Mythos to select partners due to fears of potential misuse.

Mythos: Mythos is a preview of Anthropic’s Claude AI model specialized in cybersecurity research and vulnerability discovery. It autonomously identifies and analyzes flaws in major operating systems and web browsers without domain-specific training. The news describes how its capabilities alarmed global banks, tech giants, and governments, though experts note similar detection is possible with existing models.
OpenAI: OpenAI is a prominent AI lab creating foundational models like GPT alongside specialized applications in areas such as cybersecurity. Its current models demonstrate strong capabilities in uncovering software vulnerabilities comparable to cutting-edge systems. The article quotes experts affirming that OpenAI tools can already reveal the vulnerabilities highlighted by Mythos.
Anthropic: Anthropic is an AI safety research company developing the Claude series of large language models with a focus on reliability and steerability. It recently previewed Mythos, opting not to release it publicly due to concerns over its advanced cyber exploitation potential. In this news item, Anthropic’s model triggered institutional scrambles to address exposed software risks.

Release Caution: Anthropic restricted Mythos access to select partners amid fears of misuse by criminals or adversarial actors.
Expert Consensus: Cybersecurity experts confirm that vulnerabilities uncovered by Mythos are detectable using publicly available AI models from leading labs.
Competitive Response: OpenAI introduced a limited-release cybersecurity model to address capabilities demonstrated by Anthropic’s Mythos.