Anthropic has introduced a new version of its advanced artificial intelligence system designed specifically with safety constraints in mind, responding to mounting concerns within the cybersecurity community over the dual-use potential of its most powerful model to date.
The company, known for its Claude family of AI systems, announced the launch of Fable 5 late Tuesday as a “safe” adaptation of its previously restricted model Mythos Preview. The original system had already sparked widespread concern among cybersecurity professionals after demonstrating an ability to identify critical software vulnerabilities at unprecedented speed, raising fears that similar tools in the wrong hands could significantly escalate global cyber threats.
Rather than releasing Mythos Preview broadly, Anthropic initially limited access to a small group of selected companies and governments under a program known as Glasswing, which includes around 150 participating entities. Among them are international firms and institutions, including participants in Spain, all granted controlled access in an effort to evaluate the system’s defensive applications.
Fable 5 represents the company’s attempt to extend access to a wider user base while introducing safeguards designed to limit potentially harmful applications. According to Anthropic, the model surpasses all previous publicly released systems in capability, particularly in areas such as software engineering, scientific research, and complex analytical tasks. However, the company has also acknowledged that its capabilities could be misused for activities that “could cause significant harm,” particularly in cybersecurity contexts.
To mitigate that risk, Anthropic has implemented a system in which sensitive or high-risk queries are redirected to a less capable model, Claude Opus 4.8, effectively constraining the most powerful responses in areas deemed potentially dangerous. The company describes the approach as a form of controlled capability sharing, intended to balance innovation with safety considerations.
The concerns surrounding Mythos first emerged two months earlier, when early demonstrations of the system suggested it could detect so-called zero-day vulnerabilities in software within minutes. These flaws, which are unknown to developers and therefore unpatched, typically require extensive manual investigation by expert security teams, often taking months or even years to uncover.
The implications of such speed prompted unease across multiple sectors, particularly within the financial industry, where secure digital infrastructure is essential. Cybersecurity professionals warned that if such tools were ever accessed by malicious actors, critical systems such as payment networks and customer accounts could be placed at risk.
Despite these concerns, Anthropic opted not to fully commercialize Mythos Preview, instead restricting it to controlled testing environments under the Glasswing initiative. The goal, according to the company, was to allow selected organizations to explore its defensive cybersecurity applications while preventing broader exposure.
Participants in the program, including firms specializing in cybersecurity, have since provided early assessments of the system’s capabilities. Derek Manky, a senior executive at Fortinet, one of the companies involved in the testing process, described Mythos as a highly capable but not singularly transformative tool. Speaking from Madrid, he suggested that while the model is not a “magic formula” that will redefine cybersecurity, it could still significantly alter the speed at which vulnerabilities are identified.
Manky noted that the technology could strengthen defensive operations by enabling security teams—often referred to as “blue teams”—to identify weaknesses more rapidly than adversarial actors, known in cybersecurity terminology as “red teams.” However, he also warned that if misused, such systems could potentially double the number of cyberattacks annually by accelerating the discovery and exploitation of vulnerabilities.
Other industry experts echoed a more measured interpretation of the technology’s impact. José de la Cruz, technical director for Spain at cybersecurity firm TrendAI, described Mythos as an evolutionary step rather than a revolutionary shift. He emphasized its value as a tool within existing cybersecurity frameworks, particularly in accelerating vulnerability detection and analysis.
De la Cruz noted that the system is capable of identifying security flaws and, in some cases, exploring potential exploit methods even without direct access to source code, significantly reducing the time required compared to traditional human-led analysis. He characterized it as an additional resource within a broader cybersecurity toolkit rather than a standalone transformation of the field.
The debate over the potential consequences of such systems has centered on the balance between defensive benefit and offensive risk. While proponents argue that AI-powered tools like Mythos could strengthen global cybersecurity by improving detection and response times, critics caution that the same capabilities could be replicated or adapted by malicious actors.
According to Manky, a significant concern lies in the possibility that attackers will eventually gain access to comparable systems, whether through replication or independent development. He warned that such a development could lead to a surge in the exploitation of software vulnerabilities, particularly in connected devices, where security gaps are often more widespread.
The cybersecurity industry recorded approximately 40,000 vulnerabilities in 2025, and experts suggest that advanced AI systems could substantially increase that figure if used maliciously. This potential escalation has intensified calls for stronger international cooperation in cybersecurity defense.
Industry leaders argue that collaboration across governments, companies, and security institutions will be essential to managing the risks posed by increasingly capable AI systems. Efforts to share threat intelligence and coordinate defensive strategies are seen as critical to ensuring that detection capabilities keep pace with evolving threats.
Manky emphasized the need for coordinated “blue team” alliances, designed to rapidly distribute knowledge of vulnerabilities and defensive techniques. He warned that the timeline for offensive actors gaining access to similar tools may be short, potentially within the next year, increasing the urgency for preemptive defensive preparation.
