NIST IR 8621: gaps in AI incident management

What NIST published and when

NIST's Computer Security Resource Center posted an announcement dated November 2, 2026, saying that NIST had published Internal Report (IR) 8621. The announcement says the report covers stakeholder perspectives and key findings from the May 14, 2026 workshop. It adds that the workshop brought together more than 400 participants from government, industry, academia, and other communities of interest. The event page for the workshop gives the date and place: May 14, 2026, at the NIST campus at 100 Bureau Drive in Gaithersburg, Maryland.

The announcement is a short summary of the report, and the event page carries the agenda and the invitation to the workshop. The sections below keep what each page says separate, so a reader can see which claim comes from which source.

What the report covers

The announcement says the report examines the unique challenges of managing super intelligence (SI) incidents, including cybersecurity attacks that target SI systems as well as harms that arise from misuse, malfunction, or unintended actions.

That wording gives an incident response team two separate starting points. One is an attack aimed at the system. The other is harm that comes from the system's own use or behavior, whether through misuse, malfunction, or unintended actions.

The event page uses different wording and calls the event a workshop on AI incident management. This explainer keeps each page's own term.

The event page says that AI systems are both targets and sources of risk. That matches the two starting points above: a system can be the target of an attack, and it can be the source of harm.

The event page also says the workshop would highlight AI incident types beyond cybersecurity, including misuse scenarios.

What the report says about gaps

The announcement says the report reflects stakeholder insights on gaps in preparedness and response, and on the roles and responsibilities of organizations involved in managing SI incidents.

The announcement then names areas of work, and this section takes them in turn.

The first area is guidance and developer readiness. The announcement says key areas include updating NIST cybersecurity incident response guidelines, including SP 800-61, and exploring new resources to support incident readiness among model developers. Two things follow from that sentence. Existing NIST incident response guidance is named as something to update, and model developers are named as the group for which new readiness resources would be explored.

The second area is evidence, information sharing and responsibility. The announcement says additional activities include monitoring and evidence collection, incident information sharing, and clarifying coordination and responsibilities across model developers, deployers, and users. Those items name three parts of a response: seeing and recording what happened, sharing incident information, and settling who does what when a model developer, a deployer and a user are all involved.

The announcement closes by saying these findings provide a foundation for continued stakeholder engagement.

Existing guidance named on the agenda

The agenda on the event page includes mini-talks on existing guidance. It lists talks on NIST 800-61r3, ISO/IEC 25870, and the Cyber AI Profile, NIST IR 8596. The same block lists the IQT AI Incident Response Guidebook, J-AISI IRS, and MITRE ATLAS.

For a security team, the useful point is that the guidance the workshop looked at is spread across several documents. The announcement's call to update SP 800-61 refers to one of them. The other items on the agenda list show where else stakeholders were pointed.

How the workshop was organized

The event page says the workshop would present a high-level NIST roadmap to advance AI incident response standards and practices.

The agenda's Session 1 is titled AI Under Attack: Incident Response for AI Systems, and it lists presentations from speakers at Zenity, CrowdStrike and AWS. The agenda's Session 2, chaired by Jon Boyens of NIST, lists speakers from the AI Incident Database, the Frontier Model Forum and Google Crisis Response.

The agenda splits the breakout sessions along the two starting points described earlier. They cover cybersecurity incident response for impacted AI systems, and responding to incidents resulting from misuse and malfunction of AI.

The event page lists who should attend: AI developers, service providers, incident responders, critical infrastructure partners, academics, cybersecurity professionals, and government stakeholders.

The event page says the engagement would inform work by NIST's Information Technology Laboratory and its Center for AI Standards and Innovation, including updates to existing guidelines and the development of new recommendations.

What the named areas mean for an incident response program

The announcement lists areas of work. It does not hand a security team a change list, so the points below only sort NIST's areas by the part of a program they touch.

• Written procedures. The announcement names SP 800-61 and other NIST cybersecurity incident response guidelines as material to update.
• Model developer readiness. The announcement names new resources for incident readiness among model developers, which matters to teams that depend on a developer's models.
• Monitoring and evidence. The announcement names monitoring and evidence collection as an area of work.
• Information sharing. The announcement names incident information sharing as an area of work.
• Roles and responsibilities. The announcement names coordination and responsibilities across model developers, deployers, and users.

The workshop's two incident types, attacks on AI systems and harms from misuse or malfunction, cut across all five items. A team that reads the announcement alongside the event page can check each item against both types.