Anthropic's new AI model, Fable, has sparked debate among cybersecurity researchers and professionals. The model's 'guardrails' restrict its ability to handle cybersecurity-related tasks, causing frustration among those who see it as a potential tool for enhancing security. The restrictions are in place to prevent the development of malware or biological weapons, but they have led to concerns about the model's effectiveness in a field where flexibility and adaptability are crucial. Cybersecurity veteran Matt Suiche highlights a specific issue: when asked to write secure code, Fable incorrectly identifies the task as cybersecurity-related, leading to a downgrade in its response. This highlights the challenge of balancing safety measures with the practical needs of cybersecurity professionals. The model's reliance on keyword-based guardrails is seen as a hindrance, especially when it comes to code reviews, which are essential in the cybersecurity domain. Despite these challenges, Suiche remains optimistic, suggesting that the guardrails will evolve over time as Anthropic and other AI companies collaborate with cybersecurity experts. The article also mentions Anthropic's Cyber Verification Program, which allows approved cybersecurity professionals to use Claude for more unrestricted cybersecurity work, similar to OpenAI's Trusted Access for Cyber program. The debate around Fable's guardrails underscores the ongoing tension between safety and utility in the rapidly evolving field of AI, particularly in cybersecurity.
