Algorithmic Accountability Act

Covered entities must perform and document Algorithmic Impact Assessments (AIAs) for any automated decision system or augmented critical decision process.

• Assessments must evaluate design, training data, and outcomes for bias, accuracy, and fairness.
• Entities must identify and mitigate harms related to race, color, sex, religion, age, and disability.
• Documentation must be retained and submitted to the Federal Trade Commission (FTC) upon request.

The Act targets systems that make or substantially influence consequential decisions affecting individuals' lives.

• Includes decisions related to employment, credit, housing, education, healthcare, and criminal justice.
• Covers both fully automated decisions and augmented critical decision processes where a human relies heavily on algorithmic output.
• The threshold is whether the decision has a legal, material, or similarly significant effect on the individual.

Entities must provide clear, conspicuous notice to individuals when an automated decision system is used to make a critical decision about them.

• Consumers have the right to meaningful information about how the system works and the logic involved.
• Individuals can opt out of automated critical decisions in favor of a human alternative where technically feasible.
• Adverse decisions must include an explanation and a mechanism for human review and correction of erroneous inputs.

The Federal Trade Commission is designated as the primary enforcement authority with rulemaking power.

• Entities must submit annual algorithmic impact reports summarizing assessments and mitigation efforts.
• The FTC maintains a public repository of aggregated, anonymized assessment data to promote transparency.
• Violations are treated as unfair or deceptive acts or practices under the FTC Act, carrying civil penalties and injunctive relief.

The Act mandates internal governance structures to ensure ongoing compliance.

• Covered entities must designate a responsible officer to oversee algorithmic governance.
• Requires establishment of an internal review board or equivalent mechanism for high-risk systems.
• Mandates workforce training on algorithmic bias, fairness, and the limitations of automated systems.
• Aligns with the NIST AI Risk Management Framework as a guiding standard for implementation.

The Act applies broadly to entities under FTC jurisdiction that deploy qualifying automated systems.

• Covers companies with over $50 million in annual revenue or those handling personal data of over 1 million consumers.
• Exempts small businesses and systems used solely for cybersecurity, fraud detection, or national security.
• Applies to both developers and deployers of covered systems, creating shared accountability across the supply chain.

The automated processing of personal data to evaluate, analyze, or predict aspects concerning an individual's performance at work, economic situation, health, personal preferences, or behavior.

• Forms the basis for many high-stakes automated decisions targeted by the Act
• Requires transparency about the logic involved in the profiling process
• Must be disclosed in impact assessments when used in covered systems

The legal standard requiring that a human reviewer possesses the competence, authority, and actual capacity to override an AI system's automated decision.

• Goes beyond a mere tokenistic rubber-stamping of the output
• The reviewer must have access to sufficient information to independently evaluate the decision
• A critical safeguard mandated for high-risk automated decision systems under both the AAA and EU AI Act

A mandatory, documented process for deployers of high-risk AI systems to evaluate the specific risks to the rights and freedoms of individuals likely to be affected by the system's operation.

• Directly analogous to the impact assessments required by the Algorithmic Accountability Act
• Must be completed prior to deployment and updated throughout the system lifecycle
• Requires consultation with affected stakeholders and civil society

An AI system classified under the EU AI Act as posing significant potential harm to health, safety, or fundamental rights. This classification triggers mandatory conformity assessments and ongoing compliance documentation.

• Serves as the EU's counterpart to the systems covered by the Algorithmic Accountability Act
• Requires CE marking before market placement
• Subject to post-market monitoring and serious incident reporting obligations

The immutable recording of AI-driven decisions and their inputs for auditability and the right to explanation. This technical capability is essential for demonstrating compliance with accountability legislation.

• Captures the input data, model version, and confidence scores for each decision
• Enables retrospective analysis of disparate impact and bias
• Supports the generation of meaningful explanations for affected individuals