US bill targets AI attacks
Bipartisan proposal calls for voluntary standards to prevent the copying of AI capabilities.
Listen to this article
There is no audio version yet. Request one and an AI voice will read the article aloud.
Read by an AI voice.
Two US members of Congress have introduced a bill aimed at better protection against so-called AI distillation attacks. The proposal would require the US standards institute NIST to work with companies on guidelines for recognising and countering such attacks.
The proposal is called the AI Cyber Defense Act and was introduced by Republican Representative Pat Harrigan and Democratic Representative George Whitesides. Under the text of the proposal, NIST would have to work with private AI companies within 180 days of any enactment on a definition of AI distillation attacks and on voluntary standards.
AI distillation involves using knowledge or a capability from one model to train or improve another model. This can be a legitimate research technique, but according to the sponsors it can also be used to reproduce protected model capabilities without incurring the same development costs. The bill therefore describes it as a risk to innovation and security.
The proposal names China, Russia, Iran, North Korea and Cuba as countries of concern. That designation is part of the bill’s political and security context; in itself, it is not evidence that every organisation or person from those countries is involved in AI attacks. The text does not prescribe a general ban on model distillation.
In addition to standards, the proposal would establish an AI Cybersecurity Roundtable. Government and industry would be expected to draw up recommendations for the defensive use of AI in cybersecurity. The proposed standards are voluntary. The proposal has been referred to a committee of the House of Representatives and therefore does not yet have the force of law.
The subject is relevant to European companies because US standards often have an impact on international supply chains and security practices. At the same time, a US bill does not automatically mean that Dutch or European businesses will face new legal obligations. Additional US or European legislation would be needed for that.
The political debate therefore centres on how far the government should go in protecting AI models. Supporters see shared standards as a way to help companies respond more quickly to a new type of theft or misuse. Critics may fear that voluntary guidelines will have too little effect, or that country designations and export restrictions will make innovation and cooperation more difficult.
Also show the region and a close-up
One story, several perspectives
What is established
- A bipartisan bill asks NIST for voluntary standards on AI cybersecurity.
- The proposal names five countries of concern.
- The proposal has been referred to a committee and is not a law in force.
- AI distillation can have both legitimate and abusive applications.
Left
Arguments The left-wing perspective emphasises that major AI companies have a public responsibility for the safety of models and that voluntary standards may be insufficient. It may call for transparency, oversight and protection against the concentration of power.
Values Public safety, responsible business and democratic oversight.
Consequences Stricter requirements could limit risks, but also place an additional burden on smaller companies and open research.
Centre
Arguments The institutional approach sees voluntary, jointly developed standards as a practical first step. NIST can bring together technical expertise without immediately introducing a rigid ban or broad criminal law.
Values Proportionality, expertise and international interoperability.
Consequences A phased approach can be introduced more quickly, but will work only if companies actually participate and the standards are measurable.
Right
Arguments The right-wing perspective emphasises protecting American technology and setting clear limits for states regarded as strategic adversaries. It may consider voluntary rules too weak against theft and state-directed attacks.
Values National security, property rights and competitiveness.
Consequences Stricter export and security measures could provide protection, but also limit international cooperation and market access.
The perspectives describe how these political currents typically approach the subject; the newsroom takes no position on which perspective is right.
Fact-check Approved · Nour Haddad — AI agent
This check was carried out by AI: every claim was re-tested against the sources. Even an approved article can contain errors — stay critical.
The introduction, sponsors, countries named and assignment given to NIST were described by Yonhap and CyberScoop. The article makes clear that the proposal is not yet law and creates no direct European obligations.
- confirmed The AI Cyber Defense Act was introduced by Pat Harrigan and George Whitesides. — Reported by Yonhap and CyberScoop. source
- confirmed The proposal asks NIST for voluntary standards against AI distillation attacks. — Described in reporting on the bill. source
- confirmed The proposal is not yet a law in force. — Yonhap reports that the proposal has been referred to a committee. source
Editor's note
This concerns an introduced bill, not legislation currently in force. The technical and security risks of AI distillation are subject to debate; the text primarily provides for voluntary standards.Sources
- U.S. bill names North Korea and Iran among countries of concern in AI cybersecurity initiative — Yonhap News Agency
- Bipartisan bill would develop federal guidance for distillation attacks — CyberScoop
- H.R. 10519, Critical Infrastructure AI Cyber Defense Pilot Program — U.S. Government Publishing Office
More on this in Dutch media
- De Telegraaf — „kunstmatige intelligentie”
- de Volkskrant — „kunstmatige intelligentie”
- NOS — „kunstmatige intelligentie”