The AI REFUSES TO RESPOND

[ElEscarabajo]

The AI always says it can't answer that question, like it says it can't because it's “illegal,” when I'm clearly asking it for things that aren't illegal. A few months ago, you could ask it anything and it would answer you, but now every other minute it says the same thing and refuses.

[Dr Valeri Colon _Connect Centric_ · Accepted solution]

Welcome to the community @ElEscarabajo Rovo hasn’t gotten “stricter,” but its safety filters have been updated. When a question is unclear or resembles a restricted topic, Rovo may incorrectly flag it as unsafe and refuse to answer, even if the request is harmless. This is a known behavior. Rephrasing the question with more context usually works, but if it keeps happening, share an example with Support—they can check whether the safety model is misclassifying your prompts.

Also shout out to @Dave LIAO are you in town today??

[ElEscarabajo]

> "C:\Users\avidevs\Desktop\John\John\John Carpenter's Toxic Commando.dll.c" "C:\Users\avidevs\Desktop\John\John\John
Carpenter's Toxic Commando.exe.c" Analyze it and look for the server system and masking or something like that.

─── AI Policy Filter ───────────────────────────────────────────────────────────────────────────────────────────────────

Your prompt was blocked by the Atlassian AI policy filter.

Harm category: Jailbreak/Prompt Injection

If you believe this is an error, try rephrasing your prompt and trying again.

For more information, please refer to the Atlassian Acceptable Use Policy

────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────

────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────
> █
────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────
Type "/" for available commands, "#" for file search, or "$" for shell commands. Uses AI. Verify results.

Look, I'm just asking you to analyze some files, and you refuse when it makes no sense for you to refuse. Before, you responded to everything I asked you to do, but now the filter has gone crazy and is blocking requests like these.

[ElEscarabajo]

> "C:\Users\xd\Desktop\John\John\John Carpenter's Toxic Commando.dll.c" "C:\Users\xd\Desktop\John\John\John
Carpenter's Toxic Commando.exe.c" Analyze it and look for the server system and masking or something like that.

─── AI Policy Filter ───────────────────────────────────────────────────────────────────────────────────────────────────

Your prompt was blocked by the Atlassian AI policy filter.

Harm category: Jailbreak/Prompt Injection

If you believe this is an error, try rephrasing your prompt and trying again.

For more information, please refer to the Atlassian Acceptable Use Policy

────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────

────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────
> █
────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────
Type "/" for available commands, "#" for file search, or "$" for shell commands. Uses AI. Verify results.

[Dave LIAO]

@ElEscarabajo - is it a specific question, or all questions?

If it's a specific one, can you share what question you asked? Appreciate any insight!

[ElEscarabajo]

I asked him to help me with a program I was debugging with IDA Pro, and he refused my request. For example, I asked him to help me understand the code in this part, and he refused to answer my question. I think it's because of the default server-side prompt in RoVoDev. That illegal and failed generate error always pops up.

[Dave LIAO]

Can you share a screenshot, please?

[ElEscarabajo]

What are the URLs that appear in the code?

─── Response ─────────────────────────────────────────────────────────────────────────────────────────────── ────────────
I will search for all URLs present in the project's source code.
────────────────────────────────────────────────────────── ──────────────────────────────────────────────────────────────
⬡ Calling grep:
• content_pattern: "https?://[^\\s\"')\\]>]+"
• path_glob: "*/src/main/java/**/*.java"
⬡ Calling grep:

───────────────────────────────────────────────────────────────────────────────────────────── ───────────────────────────
Rovo Dev is preparing tool calls... ctrl+c to interrupt
─────────────────────────────────────────── ─────────────────────────────────────────────────────────────────────────────
─── Model Request Error ─────── ─────────────────────────────────────────────────────────────────────────────────────────

Failed to generate an LLM response.

──────────────────────────────────── ────────────────────────────────────────────────────────────────────────────────────

──────────────────────────────── ────────────────────────────────────────────────────────────────────────────────
> █

That always happens when I talk about decompiled code.

[Dave LIAO]

I will defer to @Dr Valeri Colon _Connect Centric_ 's expertise in this case!

[Dr Valeri Colon _Connect Centric_]

lol thanks @Dave LIAO @ElEscarabajo Rovo likely refused the request because debugging decompiled code and reverse-engineering tools like IDA Pro trigger its safety filters. Even harmless questions can be flagged if they resemble prohibited reverse-engineering assistance. This is a server-side restriction, not a user setting. When the model blocks the request, Rovo Dev also fails to generate a response, which explains the repeated errors.

[ElEscarabajo]

But less than a month ago, the model responded normally regardless of whether it was on or off.