AI disclosure rule for lawmakers

Summary#

S-10692, the proposed “CANDOR Act,” would amend New York’s Legislative Law by adding a disclosure rule for remarks by state senators and Assembly members. Its stated aim is to disclose when artificial intelligence helped draft remarks; the bill is not law and is at the introduced stage.

  • A member who speaks, reads a statement, or enters comments in the official legislative record must first give a verbal disclosure if AI drafted any part of the remarks.
  • The disclosure must substantially follow the bill’s wording, including an estimate of the percentage of AI-generated text and a statement that the words are not the member’s own “organic intellect, independent thought, or cognitive labor.”
  • If a member does not comply, the bill calls for an immediate point of order and a permanent digital watermark on the remarks in the official record. The watermark would say: “WARNING: High Probability of Automated Plagiarism; Content Lacks Proven Human Origin.”

What it means for you#

  • State legislators: If the bill became law, a senator or Assembly member would have to disclose AI use before delivering covered remarks, even if AI drafted only part of them. The bill does not ban using AI to draft remarks.
  • People reading the official legislative record: Non-compliant remarks would have to be marked with the permanent digital watermark. The bill does not say who applies it or how a member could challenge or correct it.
  • Other members of the public: The bill does not directly change rules for public speech, private communications, or people outside the state Senate and Assembly.

Money#

No publicly available information on the bill’s cost is included in the supplied materials.

  • The bill does not set aside money or state a cost estimate.
  • It requires disclosures and digital watermarks in the legislative record, but does not specify staffing, technology, or other resources needed to carry this out.

What is unclear#

  • The bill does not explain how a member should calculate the “approximately” stated percentage of AI-generated text.
  • It does not set a method for deciding whether, or how much, AI drafted a remark.
  • It does not explain who would raise the point of order, apply the watermark, or handle a dispute. It specifies that the watermark is permanent but gives no correction process.
  • The bill’s definitions of AI and generative AI cover a wide range of systems, but it does not explain how they apply to common drafting or editing assistance.
  • The supplied materials do not include the current Legislative Law, so I could not verify how this rule would interact with existing legislative procedures.

Case for#

  • The bill appears intended to make AI involvement in official legislative remarks visible to people hearing them or reading the record.
  • A disclosure requirement could give the public information about how a member prepared remarks without banning AI use.
  • Applying the rule to both spoken remarks and comments entered into the official record would cover more than speeches alone.

Case against#

  • The required wording says the remarks are not the member’s own intellectual work, even when AI drafted only part. That may overstate AI’s role.
  • The bill does not explain how to measure the required percentage, which could make disclosures inconsistent or disputed.
  • The permanent watermark labels non-compliant remarks as likely plagiarism, but the bill does not define plagiarism or provide a process to contest the label.
  • The bill creates consequences for non-compliance without explaining who administers them or how they fit with existing legislative procedure.