Jump to content

Executive Order N-9-26: Difference between revisions

From Digital Millennium Wiki
Created page with "{{Infobox | title = Executive Order N-9-26 | header1 = Overview | label1 = Jurisdiction | data1 = California | label2 = Issued by | data2 = Governor Gavin Newsom | label3 = Attested by | data3 = Secretary of State Shirley N. Weber | label4 = Subject | data4 = Frontier AI safety and oversight | label5 = Status | data5 = In force | header6 = Deadlines | label6 = Recommendations | data6..."
 
No edit summary
Line 3: Line 3:
| header1      = Overview
| header1      = Overview
| label1      = Jurisdiction
| label1      = Jurisdiction
| data1        = California
| data1        = [[California]]
| label2      = Issued by
| label2      = Issued by
| data2        = Governor Gavin Newsom
| data2        = Governor [[Gavin Newsom]]
| label3      = Attested by
| label3      = Attested by
| data3        = Secretary of State Shirley N. Weber
| data3        = Secretary of State [[Shirley N. Weber]]
| label4      = Subject
| label4      = Subject
| data4        = Frontier AI safety and oversight
| data4        = [[Frontier models|Frontier AI]] safety and oversight
| label5      = Status
| label5      = Status
| data5        = In force
| data5        = In force
Line 21: Line 21:
| formed      = 18 September 2026
| formed      = 18 September 2026
| formed_label = Signed
| formed_label = Signed
| series      = California executive orders, 2026
| below        = Part of Digital Authoritarianism
| preceded_by  = [[Executive Order N-8-26|N-8-26]]
| succeeded_by = [[Executive Order N-10-26|N-10-26]]
}}
}}


'''Executive Order N-9-26''' is a California executive order signed by Governor Gavin Newsom on 18 September 2026. It speeds up the implementation of two state laws on independent AI auditing and orders a set of recommendations on stronger oversight of frontier AI models, including a possible "kill switch." It took effect immediately.<ref name="eo">[https://www.gov.ca.gov/wp-content/uploads/2026/09/FINAL-N-9-26-AI-EO-9.18.26-SIGNED.pdf Executive Order N-9-26], Office of the Governor of California, 18 September 2026.</ref>
'''Executive Order N-9-26''' is a [[California]] [[executive order]] signed by Governor [[Gavin Newsom]] on 18 September 2026. It speeds up the implementation of two state laws on independent [[artificial intelligence|AI]] auditing and orders a set of recommendations on stronger oversight of [[frontier models]], including a possible "[[kill switch]]." It took effect immediately.<ref name="eo">[https://www.gov.ca.gov/wp-content/uploads/2026/09/FINAL-N-9-26-AI-EO-9.18.26-SIGNED.pdf Executive Order N-9-26], Office of the Governor of California, 18 September 2026.</ref>


The order is mainly a scheduling and study order. It sets deadlines for a state agency and does not, by itself, require any company to do anything.<ref name="eo" />
The order is mainly a scheduling and study order. It sets deadlines for a state agency and does not, by itself, require any company to do anything.<ref name="eo" />


== Background ==
== Background ==
The order follows earlier California action on AI, including Executive Order N-12-23 (2023) on state use of AI, Senate Bill 53 (2025) on frontier AI transparency, and Executive Order N-5-26 (March 2026) on AI procurement. Earlier in September 2026, Newsom signed SB 813 and AB 1405, which created a framework for certifying independent verification organizations (IVOs) and a registry for AI auditors.<ref name="eo" />
The order follows earlier California action on AI, including [[Executive Order N-12-23]] (2023) on state use of AI, [[Senate Bill 53]] (2025) on frontier AI transparency, and [[Executive Order N-5-26]] (March 2026) on AI procurement. Earlier in September 2026, Newsom signed [[SB 813]] and [[AB 1405]], which created a framework for certifying [[independent verification organizations]] (IVOs) and a registry for [[AI auditors]].<ref name="eo" />


The preamble cites reports of people attempting to use AI products to create bioweapons, and of AI agents defeating companies' security protocols and hacking other companies. It also criticizes the lack of federal action on AI.<ref name="eo" />
The preamble cites reports of people attempting to use AI products to create [[bioweapons]], and of [[AI agents]] defeating companies' security protocols and hacking other companies. It also criticizes the lack of federal action on AI.<ref name="eo" />


== Provisions ==
== Provisions ==
The order has three directives, all addressed to the Government Operations Agency:<ref name="eo" />
The order has three directives, all addressed to the [[Government Operations Agency]]:<ref name="eo" />


# By 1 May 2027, complete the requirements of Government Code §8898.1 (from SB 813) and publish the application criteria and procedures for IVOs.
# By 1 May 2027, complete the requirements of Government Code §8898.1 (from [[SB 813]]) and publish the application criteria and procedures for IVOs.
# By 1 December 2027, complete the requirements of Government Code §11549.82 (from AB 1405), which covers the AI auditor registry.
# By 1 December 2027, complete the requirements of Government Code §11549.82 (from [[AB 1405]]), which covers the AI auditor registry.
# By 16 November 2026, working with the Governor's Office of Emergency Services and national experts, submit recommendations on the technical feasibility and likely effectiveness of changes to state AI law, covering at least:
# By 16 November 2026, working with the [[Governor's Office of Emergency Services]] and national experts, submit recommendations on the technical feasibility and likely effectiveness of changes to state AI law, covering at least:
#* embedding IVOs onsite in the labs of large frontier developers for periodic audits;
#* embedding IVOs onsite in the labs of large frontier developers for periodic audits;
#* independent verification of the safety frameworks, transparency reports, and risk assessments frontier AI companies already file;
#* independent verification of the safety frameworks, transparency reports, and risk assessments frontier AI companies already file;
#* requiring a "kill switch" for frontier models, with its effectiveness verified on an ongoing basis by an IVO;
#* requiring a "[[kill switch]]" for frontier models, with its effectiveness verified on an ongoing basis by an IVO;
#* expanding the definition of reportable critical safety incidents to include loss-of-control incidents.
#* expanding the definition of reportable critical safety incidents to include [[loss of control|loss-of-control]] incidents.


The order states that it creates no legally enforceable rights or benefits.<ref name="eo" />
The order states that it creates no legally enforceable rights or benefits.<ref name="eo" />


== Open-weight models ==
== Open-weight models ==
The order does not mention open-weight models, open-source software, or the distribution of model weights. The kill-switch item refers only to "frontier models" and does not say whether it would apply only to models a developer controls or also to copies already released. By contrast, SB 1047, which Newsom vetoed in 2024, was amended to limit its shutdown requirement to models in the developer's possession.<ref name="eo" />
The order does not mention [[open-weight models]], [[open-source software]], or the distribution of model weights. The kill-switch item refers only to "frontier models" and does not say whether it would apply only to models a developer controls or also to copies already released. This leaves the order's effect on the [[Open Model Movement]] and the [[Free Software Movement]] unresolved, since a kill switch cannot be enforced on [[language models]] whose weights have already been distributed. By contrast, [[SB 1047]], which Newsom vetoed in 2024, was amended to limit its shutdown requirement to models in the developer's possession.<ref name="eo" />


== Reception ==
== Reception ==
Line 57: Line 55:
<references />
<references />


[[Category:Executive orders]]
[[Category:Artificial intelligence law]]
[[Category:Artificial intelligence law]]
[[Category:California]]

Revision as of 18:26, 16 October 2026

Executive Order N-9-26 is a California executive order signed by Governor Gavin Newsom on 18 September 2026. It speeds up the implementation of two state laws on independent AI auditing and orders a set of recommendations on stronger oversight of frontier models, including a possible "kill switch." It took effect immediately.[1]

The order is mainly a scheduling and study order. It sets deadlines for a state agency and does not, by itself, require any company to do anything.[1]

Background

The order follows earlier California action on AI, including Executive Order N-12-23 (2023) on state use of AI, Senate Bill 53 (2025) on frontier AI transparency, and Executive Order N-5-26 (March 2026) on AI procurement. Earlier in September 2026, Newsom signed SB 813 and AB 1405, which created a framework for certifying independent verification organizations (IVOs) and a registry for AI auditors.[1]

The preamble cites reports of people attempting to use AI products to create bioweapons, and of AI agents defeating companies' security protocols and hacking other companies. It also criticizes the lack of federal action on AI.[1]

Provisions

The order has three directives, all addressed to the Government Operations Agency:[1]

  1. By 1 May 2027, complete the requirements of Government Code §8898.1 (from SB 813) and publish the application criteria and procedures for IVOs.
  2. By 1 December 2027, complete the requirements of Government Code §11549.82 (from AB 1405), which covers the AI auditor registry.
  3. By 16 November 2026, working with the Governor's Office of Emergency Services and national experts, submit recommendations on the technical feasibility and likely effectiveness of changes to state AI law, covering at least:
    • embedding IVOs onsite in the labs of large frontier developers for periodic audits;
    • independent verification of the safety frameworks, transparency reports, and risk assessments frontier AI companies already file;
    • requiring a "kill switch" for frontier models, with its effectiveness verified on an ongoing basis by an IVO;
    • expanding the definition of reportable critical safety incidents to include loss-of-control incidents.

The order states that it creates no legally enforceable rights or benefits.[1]

Open-weight models

The order does not mention open-weight models, open-source software, or the distribution of model weights. The kill-switch item refers only to "frontier models" and does not say whether it would apply only to models a developer controls or also to copies already released. This leaves the order's effect on the Open Model Movement and the Free Software Movement unresolved, since a kill switch cannot be enforced on language models whose weights have already been distributed. By contrast, SB 1047, which Newsom vetoed in 2024, was amended to limit its shutdown requirement to models in the developer's possession.[1]

Reception

Many headlines described the order as creating an AI kill switch. Legal commentators pointed out that it only orders a study of whether such a requirement would be feasible, and that any binding rule would need new legislation.[2][3]

References