MITER launches AI incident sharing initiative to improve collective AI defense

MITER's Center for Threat-Informed Defense (Center) worked with more than 15 companies to expand the community's knowledge of threats and defenses for AI-powered systems, culminating with the launch of the AI ​​Incident Sharing initiative found. The AI ​​incident sharing website and submission form are available online.

Work on the incident sharing initiative is part of the Center's Secure AI project, based on MITER ATLAS, launched in June 2024. The AI ​​Incident Sharing initiative aims to improve collective threat awareness and ultimately the defense of AI-enabled systems by enabling the rapid and protected exchange of information about attacks or accidents involving AI-enabled systems are involved. The initiative builds on two years of incident sharing collaboration across the broader MITER ATLAS community to enable faster characterization and sharing of anonymized incidents.

In parallel, the Secure AI collaboration also expanded the ATLAS threat framework to further update the adversarial threat landscape for generative AI-enabled systems. Similar to MITER ATT&CK, the ATLAS Threat Framework is a community knowledge base about attacker behavior that security experts, developers and operators use to protect AI-enabled systems.

The project added several new generative AI-focused case studies and attack techniques to the public ATLAS knowledge base, as well as new methods for mitigating attacks on AI-powered systems. MITER had previously worked with Microsoft to increase the focus of the ATLAS knowledge base on generative AI with updates released in November 2023. Through these efforts, ATLAS remains representative of the latest proven threats to AI systems in the wild.

Collaborators on the Secure AI project include AttackIQ, BlueRock, Booz Allen Hamilton, CATO Networks, Citigroup, Cloud Security Alliance, CrowdStrike, FS-ISAC, Fujitsu, HCA Healthcare, HiddenLayer, Intel, JPMorgan Chase Bank, Microsoft, Standard Chartered and Verizon store.

“As public and private organizations of all sizes and industries continue to integrate AI into their systems, the ability to manage potential incidents is critical,” said Douglas Robbins, vice president at MITER Labs. “The standardized and rapid exchange of incident information will enable the entire community to improve the collective defense of such systems and mitigate external damage.”

The MITER ATLAS AI Incident Sharing initiative provides a community of trusted contributors with protected and anonymized data on real-world AI incidents that occur in operational AI-enabled systems.

Anyone can submit an incident via the public incident sharing site. Once submitted, your organization will be considered for membership in the trusted community of data recipients. Sharing and receiving this proprietary information will enable more data-driven risk intelligence and analysis to be conducted at scale across the community.

MITER operates other public-private information sharing partnerships, including the publicly available Common Vulnerabilities and Exposures (CVE) list, which it operates on behalf of the Cybersecurity and Infrastructure Security Agency to identify, define and catalog publicly disclosed cybersecurity vulnerabilities. MITER also follows this approach with the Aviation Safety Information Analysis and Sharing (ASIAS) database to exchange data and safety information to detect and prevent aviation hazards.

Earlier this week, MITER announced the full release of the EMB3D threat model, which includes new mitigations. The advanced model helps identify threats and implement tailored security measures for embedded devices. The full public release of the EMB3D threat model is now available online. New features in the full version include tiered mitigation guidelines and alignment with ISA/IEC 62443-4-2.