Bug report: ChatGPT displays raw markup instead of rendered content.

What version of the Codex App are you using (From “About Codex” dialog)?

unknow

What subscription do you have?

plus

What platform is your computer?

windows desktop

What issue are you seeing?

Description: ChatGPT frequently outputs raw HTML-like markup, UI component tags, and citation syntax instead of properly rendered text and interactive elements.
Expected behavior: Responses should display as readable, formatted content with functioning UI elements and citations.
Actual behavior: Users see underlying markup and code, making responses difficult to read and sometimes unusable.
Impact: Disrupts readability, breaks the user experience, and persists even after requesting the response to be reformatted.

What steps can reproduce the bug?

Feedback ID: no-active-thread-01a0b090-0337-7771-9912-911a3110f275

What is the expected behavior?

Expected behavior: Responses should display as readable, formatted content with functioning UI elements and citations.

Additional information

this issue has been existing for over a week

example of the output:

4. Five questions to determine whether you really have an agent
   {@body const questions = [{q:"Does the system have a defined business objective?",d:"For example, resolve reconciliation exceptions rather than simply generate a summary."},{q:"Can it decide its next step?",d:"It can choose to investigate a duplicate, retrieve supporting evidence, or attempt an approved aggregation check."},{q:"Can it use tools to interact with its environment?",d:"For example, querying databases, calling matching functions, or creating review cases."},{q:"Can it evaluate results and adjust its approach?",d:"If an attempted match is invalid, it can investigate another possibility rather than blindly proceed."},{q:"Does it know when to stop or escalate?",d:"It respects permission boundaries, stopping conditions, and human approval requirements."}]}
   {@body const [answers,setAnswers] = DIL.useState({})}

<title size="lg">Agent qualification checklist</title> {Object.values(answers).filter(Boolean).length} / 5 {#each questions as item,i} setAnswers(prev=>({...prev,[i]:v}))} lineThrough={false}> {item.q} {item.d} {/each} {#if Object.values(answers).filter(Boolean).length===5} Agent characteristics present Your system fits the operational definition, assuming those capabilities are implemented rather than merely described. {:else} This is a design checklist, not an industry certification. The most distinguishing characteristics are adaptive decision-making and tool-directed execution. {/if} setAnswers({})}>Reset checklist

Start by reproducing the issue on the Windows desktop app using feedback ID no-active-thread-01a0b090-0337-7771-9912-911a3110f275 and compare the displayed output with the provided markup example. Done means responses render as readable formatted content with functioning UI elements and citations rather than exposing raw markup.