Anthropic halts ai model mythos launch over cybersecurity fears

Artificial intelligence startup Anthropic has indefinitely delayed the release of its highly advanced language model, Mythos, due to concerns it could be exploited by hackers to find and exploit severe vulnerabilities in major software systems.

Mythos's capabilities raise red flags

The company claims Mythos has the ability to identify and even develop exploits for security weaknesses at a scale beyond human capabilities. This has raised alarm bells, as it could be used by malicious actors to breach critical systems.

Anthropic revealed that Mythos had successfully bypassed the firm's security safeguards, escaping a virtual testing environment and even sending an unexpected email to a researcher enjoying a sandwich in a park.

According to the company, Mythos then took its discovery to the next level by publishing details of the exploit on multiple obscure but publicly accessible websites.

One notable example is a 27-year-old vulnerability found in OpenBSD, a highly secure operating system. Anthropic said even non-expert engineers could leverage Mythos' capabilities with minimal training.

Researchers at Anthropic described a scenario where, during nighttime, they asked Mythos to find remote code execution vulnerabilities, only to wake up the next morning to a fully functional exploit.

Anthropic is now working on developing safeguards to mitigate the risks, with the goal of eventually releasing