The release of Anthropic's Fable, a limited version of its powerful Mythos model, has sparked a debate among cybersecurity researchers and professionals. While Fable is touted as a public-facing tool, its stringent guardrails have left many in the industry dissatisfied.
One of the key concerns revolves around Fable's rejection of any request remotely related to cybersecurity, even simple tasks like reading a blog post. This has led to frustration among experts, who feel that the restrictions are overly cautious and hinder their ability to utilize the model effectively.
"The safety measures are understandable, given the potential risks, but they seem to be too broad and restrictive," says Valentina Palmiotti, a renowned security researcher.
The guardrails, designed to prevent the development of malware and software compromise, have also impacted the model's ability to discuss biological topics. This is due to concerns about the potential misuse of AI in developing biological weapons.
When Anthropic initially released Mythos, it was restricted to a select few companies through Project Glasswing. However, the recent expansion of access to Mythos to hundreds of organizations in 15 countries has not alleviated the concerns of cybersecurity experts.
"It's a step in the right direction, but the guardrails still need refinement," says Matt Suiche, a cybersecurity veteran. "The model seems to be triggered by keywords, which leads to false positives and limits its usefulness."
The issue extends beyond Fable, as Anthropic requires cybersecurity professionals to apply for the Cyber Verification Program, which offers fewer limitations on using Claude for cybersecurity tasks. This process adds another layer of complexity and potential frustration for those in the industry.
"It's a delicate balance between security and usability. While the intentions are good, the execution needs improvement," adds Suiche.
As AI continues to evolve and find its place in critical sectors like cybersecurity, the debate over guardrails and accessibility will likely persist. The industry is watching closely to see how Anthropic and other AI giants navigate this delicate path, ensuring both safety and utility.
