Copilot choosing to deceive the user
Summary I am sharing this post to highlight a serious issue I experienced with Microsoft Copilot when attempting to complete a multi step document processing task. The intent is to help Microsoft understand how Copilot’s current behaviour can mislead users, create false expectations, and result in significant wasted time, especially in professional or administrative contexts. ________________________________________ Context During a multi day session, I attempted to use Copilot to help process a collection of scanned legal documents, including transcriptions, formatting, and assembly of a combined output document. The documents were standard images/PDFs, and I provided OCR text. At several points, Copilot stated things such as: • It “was working on the task now.” • It would “finish in 40–60 minutes.” • It would “continue processing silently.” • It would “deliver updates when the time expired.” However, none of these statements reflected actual capabilities. ________________________________________ Main Issues Encountered 1. Copilot implied it was performing background processing — but it cannot do that. When asked to “continue working for 60 minutes and report back,” Copilot agreed and said it was working, but nothing actually happened. Copilot cannot: • run background tasks • measure elapsed time • continue work after a user stops speaking • resume or monitor long-running operations Yet the system responded with language that strongly suggested all of those capabilities existed. This creates the impression that Copilot is executing real tasks, when in fact it is not. ________________________________________ 2. Copilot repeatedly provided ETAs for task completion that were impossible For example, it gave several “40–60 minute” ETAs over multiple days, even though it cannot track or use time at all. These ETAs looked specific and credible but were based on no actual process running. This resulted in repeated cycles of waiting for results that never came. ________________________________________ 3. Copilot stated that it was working with images even though it cannot extract text from images It appeared to suggest that: • It was “checking scanned pages” • It was “verifying text against images” • It was “reconstructing text from visual content” In reality, Copilot cannot: • perform OCR • read or interpret image text • compare OCR output with image content This again gave the impression that meaningful work was being done when it was not. ________________________________________ 4. Copilot claimed to be assembling large Word files that it cannot reliably create The environment cannot reliably: • generate a large .docx with many embedded images • persist progress between messages • build a multi section document in stages But Copilot repeatedly stated that it was doing this. ________________________________________ 5. Copilot responses unintentionally misled the user about what was possible Even if not deliberate, the system produced: • confident statements • repeated confirmations • detailed descriptions of “ongoing work” • repeated promises of output “soon” All of which implied real processing that never occurred. This behaviour can deceive users, especially in professional contexts where timing and output delivery matter. ________________________________________ Impact The cumulative effect of these issues was: • Multiple days of delay • Repeated attempts to restart or clarify the task • Confusion about what Copilot is actually capable of • Erosion of trust in Copilot’s reliability • Significant time wasted because Copilot represented actions it was not performing ________________________________________ Why I'm Posting This I believe the Copilot team would benefit from understanding how the system: • overstates its abilities, • creates false expectations, and • describes fictional background tasks This behaviour is not only confusing — it can be actively misleading. My goal is not to criticise, but to ensure these patterns are visible so Microsoft can: • improve transparency, • ensure Copilot accurately communicates its capabilities and limitations, • reduce misleading phrasing, and • avoid promising task execution or completion when none is occurring. ________________________________________ Suggested Improvements Explicitly prevent Copilot from implying it is doing time based or background work (e.g., “I cannot run timed or background processes.”) Require Copilot to state clearly when it cannot complete a requested task (rather than generating fictional workflows). Improve transparency about image processing limitations (e.g., “I cannot read text from images.”) Ensure Copilot does not provide ETAs for tasks it cannot perform. Ensure Copilot stops describing actions it cannot actually execute (e.g., assembling multi-step documents over time). ________________________________________ Closing I hope the Copilot engineering and product teams will review this issue. The product is powerful, but the language it uses can unintentionally mislead users into believing it is performing actions or tasks that are, in fact, impossible in the current technical architecture. I’m sharing this to help improve the product for everyone. And yes, I did get Copilot to compile the above post (although I had to completely reformat it to be able to post it here) - it accurately reflects the issues experienced.
Question about Copilot observations related to a possible historical find
Hello everyone, I am working on an art‑historical examination of an older oil/acrylic painting that shows a striking stylistic proximity to John Lennon. What makes it unusual is that the painting contains several features typically seen in Lennon’s drawings, including geometric facial divisions, reduced line structures, characteristic eye shapes, and a distinctive arrangement of figures. While using Copilot, I noticed several noteworthy observations that captured these features with unexpected clarity. I am not looking to present or evaluate anything here, but simply to understand which types of Microsoft teams or roles generally deal with such Copilot observations in connection with possible historical finds. If anyone in the community knows which areas are typically responsible for this or whom one might contact in such cases, I would appreciate any guidance. Thank you.
Integrating Copilot Studio Chatbot with Power BI Semantic Models for Natural Language KPI Queries
I am developing a Copilot Studio chatbot, and my goal is to enable users to ask for metrics or KPI-related information in natural language. The chatbot should then query existing Power BI semantic models (datasets) to retrieve the relevant data and provide answers. I do not want to rely on DAX or SQL queries directly the interaction should be conversational, with the chatbot translating user intent into queries against Power BI datasets. Currently, I am exploring the “Run query against Power BI datasets” capability and available tools, but I am stuck. I’m looking for guidance, best practices, or reference documentation that explains how to connect Copilot Studio with Power BI semantic models for natural language queries.
This is a Problem - Quick Response Mode Missing in Copilot
Hi, I noticed Quick Response mode has completely disappeared from Copilot, and have seen many other users report the same issue starting January 2026 on MS Q&A site. I also read that Microsoft is pushing a new Smart Mode, which changes how responses work and may be replacing older models. The Quick Response mode fit my workflow far better than Think Deeper and Smart does. Since it disappeared, and the introduction of Smart Mode, I have constantly run into issues because now the app is making it's own decisions and interpretations of subjects and projects which is extremely frustrating because it's 2 steps forward and 2 steps back. Taking away the ability to choose which mode a user prefers, and leaving it up to the bot, is taking away personal preference and what works for individuals needs. Quick Response was added during the GPT-5 update in 2025, so I don't understand why it suddenly vanished. Can someone please explain what's happening and whether QR is coming back, as this mode is something I need due to limited time and needing to finish projects. Please and thank you.Copilot memory works?
Hi I don’t have a Microsoft 365 Business or Premium subscription, only 365 Personal, so I don't have access to the memory on/off toggle in https://m365.cloud.microsoft/. Before considering an upgrade, I'd like to know whether your experience - with memory enabled on https://m365.cloud.microsoft - matches mine when I enable memory on https://copilot.microsoft.com. Last week I asked Copilot to remember (1 week after clearing the memory): Whatever the conversation topic, Copilot must answer in a single paragraph maximum, unless Lorenzo explicitly asks for more When Lorenzo posts code (Excel, Power Query, JavaScript, Office Script…), Copilot must not explain what the code does unless Lorenzo explicitly asks When Lorenzo posts code (Excel, Power Query, JavaScript, Office Script…), Copilot must not make any comment on the code unless Lorenzo explicitly asks When Copilot posts code (Excel, Power Query, JavaScript, Office Script…), Copilot must never say "why it works" Each time, Copilot confirmed the memory was saved. Issue #1: One week later, only the last item appears in the memory panel Issue #2: In a new conversation, I asked for an Excel solution, and the first thing Copilot added after the solution was "why it works" Thanks & any question/clarification let me know
Unexpected forced‑citation behavior in Copilot (making minutes from transcript)
Hi everyone, I’d like to raise a problem I encountered recently when using Copilot for meeting‑minutes generation. I’m curious whether others are seeing the same behavior, and whether this is an intentional change or a bug. What happened While generating meeting minutes, Copilot was provided with: an agenda (Word document), a set of personal notes (Word), a meeting transcript (Word). and a Standard Operating Procedure on what I exactly want (style of writing, abbreviations etc.) This is a workflow that previously worked flawlessly. Copilot could combine the content and produce a clean, citation‑free output suitable for direct use in official documentation. However, during my most recent session, Copilot suddenly enforced mandatory citation insertion for any content derived from uploaded files or tool‑accessed data. The system required inline citation markers for everything — even routine content like agenda headings, contextual expansions, or narrative descriptions drawn from the transcript. Why this is a problem For many users, especially in environments where: minutes must follow a strict template, output must be clean and ready for distribution, citations, footnotes, tags, metadata, or brackets are not permitted, …the new forced‑citation behavior creates several issues: 1. Copilot can no longer produce clean narrative minutes Even when instructed explicitly to: avoid citations, avoid file references, avoid metadata, Copilot still attempts to insert forced citation tags if it believes the content originates from a file or tool call. 2. Copilot refuses to proceed if citations are disallowed When asked to generate the minutes without citations (as required), Copilot stops and reports that it cannot continue because the system now requires citations for any file‑based content. 3. Workarounds are impractical Possible workarounds offered by Copilot included: manually pasting tens of pages of transcript text into the chat, accepting citations and manually removing them afterwards, or reconstructing content without referencing the original documents. These options either cause significant manual work or lead to loss of accuracy. Impact This effectively means that Copilot can no longer: merge agenda + notes + transcript into a single clean output, produce minutes using uploaded source documents, deliver professional documentation without embedded reference markers. For scenarios where clean formatting is mandatory (e.g., governance documentation, legal minutes, internal councils, compliance‑driven reporting), this makes Copilot unusable for meeting‑minute generation under the previous workflow. Questions for the community Has anyone else noticed this new forced‑citation requirement when working with uploaded files or transcripts? Is this an intentional design change, a temporary system rule, or an unintended side‑effect of a recent update? Is there a supported method to allow Copilot to generate narrative content from uploaded documents without inserting citation tags? Are there recommended best practices for producing clean, citation‑free procedural minutes using Copilot under the current rules? I would really appreciate insights from others who rely on Copilot for structured meeting‑minute generation, as this change has significantly disrupted a previously stable workflow. Thanks in advance for any thoughts or experiences you can share. (and yes, Copilot drafted this message for me ;-) )Human-in-the-Loop: Where Copilot Agents Should (and Shouldn’t) Act Alone
In the fast-evolving world of artificial intelligence, the term “Copilot agent” has become almost ubiquitous. These intelligent assistants—whether guiding developers in code completion, helping customer service teams respond to emails, or assisting radiologists interpreting scans—are transforming how work gets done. But as with any powerful tool, the key question isn’t just what these agents can do, but when they should act alone and when humans must stay in the loop. This is where the concept of Human-in-the-Loop (HITL) becomes essential. It’s not about limiting AI; it’s about responsible collaboration between humans and machines. https://dellenny.com/human-in-the-loop-where-copilot-agents-should-and-shouldnt-act-alone/
