The Race — Building Democracy in the Age of AI. People race toward a shared finish line.
The Race

Policy tool · See record

AI incident reporting

What the tool does

Require developers or operators to report specified serious AI incidents to a responsible authority within defined deadlines. Reporting can reveal failures that would otherwise stay private and give regulators a basis for investigation and corrective action.

Current status

Enacted requirements exist, with implementation and compliance requiring separate evidence. The Illinois framework is a retained state-law anchor; California also has critical-incident reporting duties. A federal proposal should not be represented as an enacted nationwide requirement.

Where it has been used

Illinois’ AI incident-reporting provisions

Added by

Illinois’ Artificial Intelligence Safety Measures Act is identified as an incident-reporting example alongside annual audits, whistleblower protections and public frontier-risk frameworks. The available account does not establish the detailed current commencement and reporting terms, so it should not be treated as confirmation that every duty is already operating.

Read more about this case →

California’s critical-incident reporting

Added by

California’s September 16, 2026 announcement describes SB53’s critical-safety-incident reporting requirements. The mechanism requires covered developers to report serious safety events rather than keeping all information inside the company. Adoption and actual compliance are different stages.

Read more about this case →

New Mexico’s proposed incident duties

Added on

New Mexico’s proposed frontier-safety legislation describes 24-hour and 72-hour incident-reporting duties. It is intended legislation for 2027, not a compulsory reporting system already in force. The complete coverage, triggers and exceptions are not established by the public summary alone.

Read more about this case →

GSA’s supplier incident-reporting clause

Added on

GSA adopted a standard contract clause on September 28 for government purchases in which a large language model is a material feature and directly receives or produces government data. Mandatory use in new covered purchases starts October 19; contracting officers may use it earlier. Incorporating it into an existing contract is a separate contracting decision. This is an adopted purchasing rule, not a single new supplier contract that every government can immediately order from.

The clause protects covered government inputs, outputs and custom developments; limits their reuse for training, sale and marketing; requires portable data and documented interfaces; and protects migration from obstructive charges. It includes reasonable human intervention and action traces, incident reporting within 72 hours, ordinarily 30 days' notice of material changes, and suspension and noncompliance remedies. Supplier ownership of pre-existing models and technology remains. Decommissioning liability after qualifying termination is capped at 25% of the affected order; replacement-system and reprocurement costs are excluded.

Eligible state, local and tribal governments and public educational institutions can use relevant GSA purchasing programmes within their statutory scope. If the underlying eligible contract contains this clause and the supplier accepts the non-federal order, its applicable terms carry into that separate purchase. Suppliers may reject those orders; eligibility, product category and incorporation must be checked for the particular purchase. The protection is therefore potentially reusable well beyond federal agencies, without assuming universal coverage or automatic amendment of all existing contracts.

Read more about this case →

Cantwell’s proposed reporting framework

Added on

Senator Maria Cantwell published a governance framework October 7. It calls for measurable catastrophic-risk standards, independent pre-release audits, disclosure and incident reporting, accountability and whistleblower protections, public defensive capacity, and human oversight and recourse. It also addresses children, workers and international coordination.

This is an outline of desired legislation, not an enacted system or a newly enforceable obligation. Its value is that it describes several concrete protections that could replace the vague instruction to “regulate AI.” It does not yet specify all the institutions, thresholds, remedies or implementation details needed to make those protections work.

Read more about this case →

Pennsylvania’s proposed incident reports

Added on

Pennsylvania House Bill 2829 was introduced and referred to committee October 5. It targets developers with more than $500 million in annual revenue and frontier models meeting a specified training-computation threshold. If enacted, its disclosure and filing system would begin January 1, 2028.

The bill calls for a documented safety framework, annual independent audits, auditor access to unredacted information, incident reports and whistleblower protections. The Pennsylvania Emergency Management Agency would administer the system, and the attorney general could seek penalties and injunctions. Its catastrophic-risk definition is narrow: it focuses on specified very large harms, rather than creating comprehensive protection against every harmful AI use.

Read more about this case →

Britain’s AI security-practice agenda

Added on

Britain’s AI Security Institute has published the agenda promised in the minister’s October 12 speech. It asks developers and technical organizations to collaborate on practices covering the entire development process, including training, internal deployment and evaluation. Its priorities include risk documentation, monitoring, incident reporting, evaluator access/resources/publication, and measurement of dangerous capabilities.

The institute explicitly says it is not mandating requirements. It proposes developing and sharing voluntary practices first, with mature practices potentially passing into international standards organizations later. No adopted compulsory standard, enforceable incident-reporting duty or new public right of access is established by this publication. It supplies more detail than the speech, while leaving authority and participation terms to future arrangements.

Read more about this case →

Scientists’ international AI security proposal

Added on

On October 13, 2026, an international group including Geoffrey Hinton and Yoshua Bengio called for stronger containment of AI systems capable of cyber operations. The proposal would initially isolate such systems from external networks, develop technical standards and create an international body for incident information and oversight.

An air gap means disconnecting a test environment from the internet or other networks. The letter asks governments and developers to cooperate; its publication does not establish an operating regulator, an adopted reporting obligation or a funded international institution.

Read more about this case →

Federal agencies publish voluntary practices for serious frontier-AI incidents

Added on

NIST, NSA and CISA published ten voluntary practices on October 16 for incident readiness and response during advanced-model training and evaluation. The joint fourteen-page document addresses events such as unexpected agent activity or loss of control: prepare technical and personnel capabilities, monitor systems, preserve evidence, contain the incident and check containment before restarting. It also recommends coordination and information sharing, including notification to government.

These are voluntary operational recommendations. They do not create a mandatory public incident-reporting duty, an independent inspection power or a demonstrated compliance result. The document’s reference to super-intelligence training is a preparedness scope, not evidence that such a system currently operates.

Government endorsement of voluntary practices can lend legitimacy to company-centered oversight. These practices nevertheless preserve applicable legal reporting requirements and recommend external evaluation and public sharing. They do not establish that industry oversight has replaced public enforcement.

Read more about this case →

Source snapshot: 2026-10-22. Individual event and verification dates remain as recorded above. Inclusion alone does not assign a score.