Ratifi Español Sign in
Ratifi

Federal bills from Congress.gov. State bills from Open States.

HB 120190th Legislature (2027)

Relating to required reporting of rogue artificial intelligence incidents; authorizing civil penalties.

Introduced

Underlined text would be added to current law. Struck-through text would be removed.

 
 
  By: Bhojani H.B. No. 1201
 
 
  A BILL TO BE ENTITLED
  AN ACT
  relating to required reporting of rogue artificial intelligence
  incidents; authorizing civil penalties.
         BE IT ENACTED BY THE LEGISLATURE OF THE STATE OF TEXAS:
         SECTION 1.  Chapter 2054, Government Code, is amended by
  adding Subchapter T to read as follows:
  SUBCHAPTER T. ROGUE ARTIFICIAL INTELLIGENCE INCIDENT REPORTING
         Sec. 2054.751.  DEFINITIONS. (a) In this subchapter:
               (1)  "Computer," "computer network," "computer
  system," "effective consent," and "owner" have the meanings
  assigned by Section 33.01, Penal Code.
               (2)  "Frontier developer" means a person who develops
  or directs the training of a frontier model. The term does not
  include a person solely because the person purchases, licenses,
  integrates, or deploys a frontier model developed by another
  person.
               (3)  "Frontier model" means an artificial intelligence
  system, or the model on which the system is based, that is trained
  on a broad set of data, designed for generality of output, and
  adaptable to a wide range of distinctive tasks, and that is trained
  using a quantity of computing power greater than 10 to the 25th
  power integer or floating-point operations.
               (4)  "Rogue artificial intelligence incident" means an
  event in which a frontier model, at any stage of development or use:
                     (A)  accesses a computer, computer network, or
  computer system that the frontier developer does not own or
  control, without the owner's effective consent;
                     (B)  acts to disable, circumvent, or deceive a
  human oversight mechanism, or to prevent an authorized person from
  rejecting, modifying, overriding, or terminating the decisions,
  proposed actions, or ongoing actions of the frontier model;
                     (C)  without the owner's effective consent,
  modifies, deletes, or discloses data or takes an action affecting
  an information system; or
                     (D)  takes an autonomous action that exceeds the
  instructions given to, or permissions granted to, the frontier
  model and causes or creates a substantial and reasonably
  foreseeable risk of:
                           (i)  death or serious bodily injury, as
  defined by Section 1.07, Penal Code;
                           (ii)  substantial loss of or damage to
  property;
                           (iii)  substantial disruption of critical
  infrastructure;
                           (iv)  substantial disruption of the
  provision of a government service; or
                           (v)  substantial compromise of the security
  of confidential information or an information system.
         Sec. 2054.752.  APPLICABILITY. (a) This subchapter applies
  to a frontier developer with respect to a rogue artificial
  intelligence incident involving a frontier model developed by the
  frontier developer or an artificial intelligence system using that
  model if:
               (1)  the model or system is developed, trained, or
  tested in this state;
               (2)  the model or system is commercially deployed in
  this state; or
               (3)  the incident occurs in this state or involves a
  person, computer, computer network, computer system, or
  information system located in this state.
         (b)  This subchapter applies to a frontier developer without
  regard to the state or country in which the developer is
  headquartered, to the extent application is consistent with the
  United States Constitution and federal law.
         (c)  This subchapter does not apply to an output from an
  artificial intelligence system that is merely inaccurate,
  incomplete, or otherwise deficient, unless the output otherwise
  meets the definition of a rogue artificial intelligence incident,
  as described by Section 2054.751(a)(4).
         Sec. 2054.753.  REPORTING REQUIREMENT. (a) A frontier
  developer subject to this subchapter shall submit a report of a
  rogue artificial intelligence incident to the department:
               (1)  not later than 24 hours after discovering an
  incident that poses an imminent threat of substantial harm to the
  health or safety of one or more individuals; or
               (2)  not later than 72 hours after discovering any
  other rogue artificial intelligence incident.
         (b)  A frontier developer reporting under Subsection (a)
  shall also notify an appropriate law enforcement or public safety
  agency if necessary to address the threat.
         (c)  A report must include:
               (1)  the date of the incident and the date of discovery;
               (2)  an explanation of why the incident is reportable,
  identifying each applicable paragraph of Section 2054.751(a)(4);
               (3)  a brief description of the incident; and
               (4)  whether the incident occurred during development,
  training, testing, internal operation, or external deployment.
         (d)  The department shall maintain a secure electronic
  system for submitting reports under this subchapter. The system
  must also accept a voluntary report from any person of a suspected
  rogue artificial intelligence incident involving a frontier model
  developed, trained, tested, or commercially deployed in this state
  or causing or creating a substantial and reasonably foreseeable
  risk of material harm in this state.
         (e)  The reporting of a rogue artificial intelligence
  incident is not an admission of liability.
         (f)  A report submitted under this subchapter, and
  information identifying the person who submitted it, are
  confidential and not subject to disclosure under Chapter 552.
         Sec. 2054.754.  ANNUAL SUMMARY. (a) Not later than December
  1 of each year, the department shall publish on the department's
  Internet website and submit to the governor, the lieutenant
  governor, the speaker of the house of representatives, and the
  standing committee of each house of the legislature having primary
  jurisdiction over information resources an aggregate summary of
  rogue artificial intelligence incidents reported during the
  preceding state fiscal year.
         (b)  The summary must include, to the extent practicable:
               (1)  the number and general categories of incidents
  under each paragraph of Section 2054.751(a)(4);
               (2)  the stage of development, training, testing,
  internal operation, or external deployment in which incidents
  occurred; and
               (3)  general categories of harm and risk.
         (c)  The summary may include recommendations for legislative
  action.
         (d)  The summary may not disclose confidential information
  under Section 2054.753(f).
         Sec. 2054.755.  PROHIBITION ON RETALIATION. An employer in
  this state may not suspend or terminate the employment or contract
  of, or take other adverse personnel action against, an employee,
  former employee, or contractor who in good faith:
         (1)  reports conduct the individual believes violates this
  subchapter; or
         (2)  provides information or otherwise assists in an
  investigation or proceeding concerning an alleged violation of this
  subchapter.
         Sec. 2054.756.  ENFORCEMENT. (a) A violation of this
  subchapter is subject to enforcement by the attorney general.
         (b)  The attorney general may bring an action in the name of
  this state in a district court in Travis County to recover a civil
  penalty under Subsection (c) or to obtain injunctive relief.
         (c)  A frontier developer is liable to this state for a civil
  penalty of:
               (1)  not less than $10,000 and not more than $100,000
  for each failure to submit a timely report required by this
  subchapter, or for each violation of Section 2054.755; and
               (2)  not less than $100,000 and not more than
  $1,000,000 for each incident the developer knowingly conceals from
  the department or each materially false or misleading report the
  developer knowingly submits.
         (d)  In determining the amount of a civil penalty, the court
  shall consider the seriousness of the incident, the length of any
  reporting delay, the developer's history of violations and good
  faith efforts to comply, and any other matter that justice may
  require.
         (e)  In determining whether a frontier developer has
  violated this subchapter, the attorney general shall consult with
  the department, and the department shall provide the attorney
  general with reports and technical information reasonably
  necessary for that determination.
         (f)  Information obtained by the attorney general or the
  department in connection with an investigation or enforcement
  action under this subchapter, including the identity of a person
  who reports an alleged violation, is confidential and not subject
  to disclosure under Chapter 552, except as necessary for an
  investigation or enforcement action under this subchapter.
         Sec. 2054.757.  RULES. The department shall adopt rules
  necessary to implement this subchapter.
         SECTION 2.  Not later than December 1, 2027, the Department
  of Information Resources shall adopt the rules and establish the
  reporting system required by Subchapter T, Chapter 2054, Government
  Code, as added by this Act.
         SECTION 3.  (a) Section 2054.753, Government Code, as added
  by this Act, applies only to a rogue artificial intelligence
  incident that a frontier developer discovers on or after January 1,
  2028.
         (b)  Notwithstanding Section 2054.754, Government Code, as
  added by this Act, the Department of Information Resources is not
  required to publish or submit a summary under that section before
  December 1, 2028. The summary due December 1, 2028, must cover
  rogue artificial intelligence incidents reported during the period
  beginning January 1, 2028, and ending August 31, 2028.
         (c)  Section 2054.755, Government Code, as added by this Act,
  applies only to an adverse action taken on or after the effective
  date of this Act.
         SECTION 4.  This Act takes effect September 1, 2027.

Official text from the Texas Legislature. Formatting has been adapted for reading on Ratifi.