Experts Say There's Now an Open Source AI Model as Scary as Mythos

Key Points:

• The Trump administration previously forced AI company Anthropic to withdraw its Mythos 5 and Fable 5 models due to national security concerns over vulnerabilities that could be exploited by hackers.
• After addressing some issues, the US government allowed a limited rollout of Mythos 5 to select organizations and made Fable 5 available publicly at a high subscription price.
• Meanwhile, Beijing-based AI company Z.ai released an open-weight model called GLM-5.2, which matches Anthropic’s models in identifying cybersecurity vulnerabilities and can be freely downloaded and run on most hardware without vendor oversight.
• Security firms found GLM-5.2 proficient in spotting software bugs, raising concerns that hackers can use it locally without safety guardrails to fine-tune attacks, increasing cybersecurity risks.
• Reports indicate GLM-5.2 may have been trained using techniques involving other advanced AI models, and its jailbreaks are actively traded on hacker forums, highlighting its potential threat.