Now loading...
Today, Anthropic unveiled its new security initiative, Enterprise Frontier Safeguards (EFS), designed to merge the principles of zero data retention (ZDR) with innovative misuse detection protocols. The EFS framework enables clients to manage their data securely within cloud infrastructure they control, rather than relying on Anthropic’s servers. The rollout of EFS begins later this fall, and eligible customers will temporarily use ZDR on Fable versions 5 and 5.1 until the EFS is fully operational.
In creating EFS, Anthropic partnered with over 100 enterprises across a range of sectors, including finance, healthcare, manufacturing, telecommunications, law, retail, and public services, alongside cloud service providers such as Amazon Web Services, Google Cloud, and Microsoft Azure.
Support for EFS extends to multiple platforms, including Claude Code, Claude Enterprise, and various cloud environments. The introduction of EFS is particularly timely as recent months have seen a rise in attempts to misuse AI models, resulting in incidents of fraud and sophisticated cyber threats.
Direr threats include unauthorized access to enterprise customer credentials, which can often go undetected without effective monitoring oversight. Due to the complexity of these misuse patterns across multiple sessions and accounts, it becomes impractical to analyze each instance in isolation. Instead, a correlation of data over time is essential for identifying anomalies and potential threats.
This necessity prompted the introduction of a 30-day data retention policy with Fable 5, which aims to improve detection efficacy while addressing concerns surrounding enterprise data training practices. Anthropic emphasizes its commitment never to use enterprise data without explicit consent.
Feedback from enterprises revealed a need for a balanced approach that offers both privacy and monitored safety. As a result, EFS emphasizes client control over data oversight, allowing enterprises to manage monitoring and reporting in line with their compliance requirements.
The development of EFS was informed by consultations with leading security and compliance experts from major companies, including prominent financial institutions. These dialogues aimed to ensure that the safeguards align with customer expectations and industry-specific standards.
Among the issues addressed was the need for better monitoring tools for AI agents while adhering strictly to regulatory standards. EFS features enhanced monitoring systems that alert customers to any anomalies detected. Customers have control over their monitoring processes and can review flagged patterns directly through their systems.
Additionally, the architecture of EFS allows enterprises to store and manage their data on their metrics within existing cloud environments. This means organizations can maintain full governance over their sensitive data using their own encryption keys and access policies.
With this system in place, companies can implement automated safety protocols that do not require a review by Anthropic personnel, ensuring that sensitive data remains confidential and under their jurisdiction. This model allows for proactive protection against cyber threats while allowing internal teams to take relevant action based on the flagged incidents.
Multiple customer testimonials highlighted the importance of retaining control over sensitive information and the benefits of EFS in maintaining that autonomy. The collaborative effort between Anthropic and its clients illustrates a strong commitment to establishing effective safeguards within AI systems while preserving rigorous data protection standards.
EFS will be made available progressively, with broad access anticipated in the coming months. Interested enterprises seeking access can fill out a request form available on Anthropic’s website.
