NIST to set AI cybersecurity standards

Summary#

The AI Cyber Defense Act is a bill that would add duties for the National Institute of Standards and Technology (NIST); it does not amend an existing Act. Its stated goal is to develop voluntary standards for responding to certain AI attacks and using AI for cybersecurity. The bill was introduced in the House and referred to committee; it is not law, and passage is uncertain.

  • Within 180 days of enactment, NIST’s Director must seek to engage private-sector AI stakeholders to establish a definition of AI distillation attacks and voluntary standards and best practices to detect and prevent them.
  • NIST must publish those standards and best practices on its website within one year after they are established.
  • Within 180 days of enactment, the Director must establish an AI Cybersecurity Roundtable to develop voluntary standards and best practices for using AI defensively for cybersecurity.
  • The Director must seek representation on the roundtable from private-sector AI stakeholders and owners and operators of critical infrastructure. NIST must publish the roundtable’s standards and best practices within one year after the roundtable is established.
  • The bill excludes certain “entities of concern” from its definition of private-sector AI stakeholders. The Director may also determine that an entity is of concern.

What it means for you#

  • NIST: The Director would have to seek industry engagement, establish the roundtable, and publish the resulting voluntary guidance by the bill’s deadlines.
  • AI businesses and critical infrastructure operators: NIST must seek engagement with private-sector AI stakeholders, and the Director must seek representation from those stakeholders and infrastructure operators on the roundtable. The bill does not require any business to join, follow the guidance, or change its practices.
  • People and organizations generally: The bill does not create a new right, ban, fine, or required cybersecurity service. It creates a roundtable within NIST, not a new independent agency.

Money#

No cost information is in the available material.

  • The bill would require NIST to carry out these activities, but it gives no cost estimate, staffing level, or separate funding.
  • It creates no fee, fine, tax, or payment to businesses.

What is unclear#

  • The bill does not define “AI distillation attacks”; it directs NIST to seek to establish a definition with private-sector stakeholders.
  • It does not say how NIST must choose roundtable members, or what happens if it cannot secure representation from a group.
  • The Director may designate an entity as an “entity of concern,” but the bill gives no criteria or review process for that decision.
  • The bill sets deadlines for establishing and publishing guidance but does not set rules for updating it or measuring whether it is used.
  • The bill refers to definitions in other laws, but those laws were not included in the supplied material, so those definitions could not be checked here.

Case for#

  • The bill appears intended to help address AI-related cyber risks by asking NIST to develop guidance with input from AI businesses and critical infrastructure operators.
  • Voluntary guidance could give organizations a shared starting point without requiring every organization to adopt the same practices.
  • A separate roundtable on defensive uses of AI could bring together people who develop AI and those responsible for critical infrastructure.

Case against#

  • The standards are voluntary, so the bill does not ensure that organizations adopt or follow them.
  • The bill leaves the definition of AI distillation attacks to a later process, so the scope of that work is not yet clear.
  • The Director’s power to designate an entity of concern has no stated criteria or review process in the bill.
  • The bill requires NIST to do this work but provides no funding or staffing estimate; the resources needed are unknown.