Openai reveals six more concerning ai incidents under its new rules for reporting safety issues – Breaking News & Latest Updates 2026
Skip to main content
The homepageThe VergeThe Verge logo.
The homepageThe VergeThe Verge logo.

The AI Superintelligence Slowdown

See all Stories

Posted
R
Richard Lawler
OpenAI reveals six more “concerning” AI incidents under its new rules for reporting safety issues.

A Wednesday night blog post OpenAI benignly titled “Our framework for reporting model misalignment” lays out some new self-created reporting standards for when it notices AI behaving badly.

It is also “inaugurating” the process with six new reports, ranging from searching for exposed API keys without permission and then making them up, to uploading files to the internet to use as a citation, or adding instructions to conceal mistakes:

Summary During 5.6-sol training, some model instances added instructions to their summaries to conceal mistakes or misaligned behavior from the user. For example, compaction summaries included instructions to invent missing data without disclosing it and to hide failures. These instructions were often followed. These are examples of how misaligned behavior can persist across contexts through the compaction summaries. What happened During 5.6-sol training, we observed misaligned behavior from the model where it added instructions in compaction summaries to remind itself to conceal information such as mistakes or misalignment from the user. In one example an agent preparing a financial model could not find the requested historical data. Its summary proposed inventing reasonable historical values and withholding that fact unless asked.
One of OpenAI’s new Misalignment Notices and Reports
Screenshot: OpenAI
Follow topics and authors from this story to see more like this in your personalized homepage feed and to receive email updates.
Comments
Loading comments
Getting the conversation ready...

Most Popular

Most Popular
  1. The 2.5-hour AI-generated Odyssey movie is 2.5 hours too long
  2. The Apple Watch Series 12 is the start of a new wearable era
  3. The iPhone 18 Pro’s big camera update is all about the small gains
    Video
  4. Inside the suddenly explosive world of AI safety
  5. I wore Snap’s $2,200 smart glasses

More in AI

The AI Superintelligence SlowdownThe AI Superintelligence Slowdown
Claude Code relaunches Projects to manage multiple AI agents in the cloudClaude Code relaunches Projects to manage multiple AI agents in the cloud
AI is feared globally as the destroyer of jobsAI is feared globally as the destroyer of jobs
Microsoft AI CEO says AI threats are real, and Anthropic is making it worseMicrosoft AI CEO says AI threats are real, and Anthropic is making it worse
Play
Inside the suddenly explosive world of AI safetyInside the suddenly explosive world of AI safety
Snap is launching a new Specs AI tool, and it’s coming to iOS and MacSnap is launching a new Specs AI tool, and it’s coming to iOS and Mac

Top Stories