US Cybersecurity Experts Warn Trump Administration's AI Stance Risks Weakening Defenses
Cybersecurity leaders and AI researchers are expressing concerns that the Trump administration's actions regarding access to Anthropic's Fable 5 and Mythos 5 AI models could weaken U.S. cyber defenses. They warn that setting a precedent discouraging American AI companies from developing tools to identify and fix vulnerabilities may give adversaries an advantage. Nearly 150 security leaders signed an open letter, organized by former Facebook security chief Alex Stamos, urging the administration to reverse its restrictions. The dispute arose following Amazon's reported concerns about a 'jailbreak' capability in Fable, which security experts suggest is a common feature used for defensive security research.
A growing number of AI researchers and cybersecurity leaders have voiced fears that the Trump administration is establishing a precedent that could deter American AI companies from creating tools essential for identifying and remediating cybersecurity vulnerabilities. These experts warn that efforts by the administration to prevent an AI hacking crisis may inadvertently weaken U.S. cyber defenses.
Former Facebook security chief Alex Stamos stated that the administration has created a precedent where American models are restricted from conducting defensive security research. Stamos organized an open letter, co-signed by nearly 150 security leaders, calling on the Trump administration to reverse its decision to restrict access to Anthropic's Fable 5 and Mythos 5 models.
The administration's actions reportedly stemmed from concerns about potential Chinese access to Mythos and a call from Amazon CEO Andy Jassy. Amazon's concerns reportedly centered on a 'jailbreak' identified in Fable, which allows the model to generate 'proofs of concept'—a capability often utilized by security teams to comprehend and resolve vulnerabilities.
Anthropic, in response, stated that it had collaborated with internal teams and external security researchers to test Fable 5 for flaws prior to its release. The company maintains that achieving 'perfect jailbreak resistance is not currently possible for any model provider' and has focused on making jailbreaks narrow or costly to produce. Katie Moussouris, CEO of Luta Security, also reviewed Amazon's findings, stating the issue involved prompts supporting defensive security work rather than mass exploitation.
Cyber experts are concerned that if AI companies fear repercussions for developing models capable of identifying vulnerabilities, they might remove functionalities crucial for defenders. Researchers argue that this response risks granting an advantage to adversaries, as Chinese AI developers and government-backed hacking groups are unlikely to abandon similar tools. Stamos noted that this situation could be perceived as a significant threat to American innovation.
The U.S. government is in the process of establishing a vulnerability clearinghouse under a recent AI security executive order, intended to triage reports regarding AI model threats such as jailbreaks and prompt injections. However, questions remain concerning the amount of cybersecurity talent within the Trump administration following recent White House departures and the reduced role of the nation's leading cyber agency. (Source: Axios)
