Production AI Institute · Public record

Production AI Institute · Public record of production AI

What changed in production AI, backed by evidence you can cite.

Source-backed records of vendor changes, incidents, and disclosures - mapped to an open production standard so teams can inspect what happened and what it means.

Open today's briefingCheck an AI tool
Built by the InstituteOpen work · Public evidence
Company operating guideAI Adoption GuideFive stages from gated access to safe, AI-native operations.
ShareShare on XShare on LinkedIn
Latest production change

Newest observed change

Anthropic: Claude Opus 4 retired

Claude Opus 4 has been retired. The retirement date was not fully captured in the snapshot but the table row indicates a retired state.

17 July 202680% confidence1 source
Anthropic
Read the full record →

·

68Production changes
4Policy editions
1556Harm records indexed
4132Sources on file
Production changes

What changed in the stack

Vendor outages, policy shifts, disclosures, and Lab signals PAI observed and mapped to the standard.

Release
Anthropic: Claude Opus 4 retired
D3D417 July 202680%1 source
Claude Opus 4 has been retired. The retirement date was not fully captured in the snapshot but the table row indicates a retired state.
→
Incident
Supabase: S3 endpoints for keys with special characters broken
D4D816 July 202692%2 sources
We have identified an issue where S3 endpoints for keys with special characters will be broken. We are actively investigating the correct mitigation for this issue.
→
Incident
Disruption with some GitHub services
D4D816 July 202692%2 sources
The degradation has been mitigated. We are monitoring to ensure stability.
→
Incident
Anthropic: Elevated errors for Claude Opus 4.7
D4D816 July 202692%2 sources
We are currently investigating this issue.
→
Incident
OpenAI: Elevated Error Rates For SSO Login
D4D816 July 202692%1 source
We have applied the mitigation and are monitoring the recovery.
→
Harm records

Incidents on the public index

Harm rows indexed from the AI Incident Database, then mapped to PSF controls for operators.

Incident
blockclubchicago.org: Coco Delivery Robot Reportedly Crashed into Chicago Bus Shelter
D4D5AI Incident Database · Incident 15684 July 202678%2 sources
A Coco delivery robot reportedly collided with a glass bus shelter in Chicago's Old Town neighborhood, shattering panels and scattering glass across the sidewalk. Coco called the crash a rare incident, said no one was injured, and accepted responsibility for cleanup and repair costs while investigating what caused the robot to strike the structure.Operator lens (high): inspect D4 (observability), D5 (model change gate).
→
Incident
bbb.org: Purported Deepfakes of Public Figures and Celebrities Reportedly Used to Market Honey-Based Supplements as Alzheimer's Treatments
D2D7AI Incident Database · Incident 15642 July 202682%3 sources
Beginning around October 2025, a campaign reportedly used AI-generated videos and cloned voices of well-known public figures and celebrities such as Bill Gates, Sanjay Gupta, Anderson Cooper, Bruce Willis, and others to market honey-based supplements as treatments for Alzheimer's disease and dementia. Consumers reportedly said modest purchases sometimes led to continuing charges or unwanted shipments, while cancellation and refund requests were difficult to resolve.Operator lens (medium): inspect D2 (output validation), D7 (tool permissions).
→
Incident
gptzero.me: KPMG Report Allegedly Included Hallucinated Citations and Inaccurate Agentic AI Case Studies
D2D6AI Incident Database · Incident 156330 June 202678%2 sources
In October 2025, KPMG published "Total Experience: Redefining Excellence in the Age of Agentic AI." GPTZero later reported that only five of 45 citations accurately matched their sources, while others contained flawed details. The pattern reportedly suggested an unverified LLM research tool, but the system and extent of AI involvement remain unconfirmed.Operator lens (medium): inspect D2 (output validation), D6 (autonomy oversight).
→
Incident
reuters.com: Kalibrate Fuel Pricing Allegedly Enabled California Gas Station Operators to Coordinate Higher Pump Prices
D2D6AI Incident Database · Incident 155929 June 202682%3 sources
Major gas station operators in California allegedly used Kalibrate Fuel Pricing to automate pump prices using competitor data and reduce price competition. A June 2026 complaint said the AI-based system increased gasoline prices by as much as 30 cents per gallon where adoption was widespread.Operator lens (medium): inspect D2 (output validation), D6 (autonomy oversight).
→
Incident
queensda.org: Former New York City Council Candidate Jonathan Rinaldi Allegedly Used AI to Create False Endorsements and Fabricated News Reports
D2D6AI Incident Database · Incident 155725 June 202678%2 sources
Queens prosecutors alleged that former City Council candidate Jonathan Rinaldi used AI during his 2025 campaign to manufacture the appearance of political support through posts made to resemble authentic endorsements and news coverage. The material reportedly attributed backing to public figures and institutions that had not provided it and falsely claimed that opponent Lynn Schulman had withdrawn. Rinaldi was charged with forgery and possession of forged instruments.Operator lens (medium): inspect D2 (output validation), D6 (autonomy oversight).
→
Continue on today's briefing →

Also from the Institute

The Institute is more than the briefing

The record is the spine. These open instruments turn evidence into standards, independent evaluation, and practical action.

The Framework

The open production safety framework every record maps to.

8 domains

The Lab

Independent model and agent scorecards against the Framework.

5 published

AI Morality Compass

Public moral-reasoning cases and evaluations across leading models.

12 cases

AI Adoption Guide

Five operating stages for scaling AI agents with evidence and control.

5 stages

Keep going

Public record

This record is maintained by PAI and free to cite. If something is wrong or missing, tell us. Corrections and source suggestions keep the record honest.

Follow policy changes ->Save a watch ->Submit a correction
Records are free to cite. citation guidance.