pls fix this @OpenAI_Support or at least explain what’s going on
Please @OpenAI_Support, respect your users. As a reminder, most of your users with access to Daybreak Blue use Daybreak for their professional work and pay for high-tier plans to use it. At the very least, some communication from you acknowledging that we are seen and heard would reassure us that we haven’t simply been forgotten.
We are specialists in our fields, and we provided evidence of that expertise to qualify for your program. You gave us the opportunity to do our work with your tools, so respect us.
I pay $240 per month, and even more after conversion to euros. The work I started with your tools is now on hold, with no communication from you. Take the necessary steps to address this.
Yes, hopefully this bug (I hope so, or they intend it) can be fixed quickly. So, I will turn to Claude Code’s CVP. I just tested it, and it works pretty well and won’t block my request.
At this point, it seems like OpenAI wants cybersecurity practitioners to switch to Anthropic. In just the past month, they’ve halved usage limits on the Pro 200 plan, made Codex Security Cloud a paid service, and modified the guardrails on Daybreak Blue. Once this billing cycle ends, I will be cancelling my subscription. Anthropic seems to have better models, better usage limits, and now less restrictive guardrails.
After watching this thread the common denominating factor seems that you have Pro subscriptions, not sure if this is the case for everybody in this thread however…?
I’m on Plus and have Daybreak Blue but I am not experiencing this issue at all.
SAME HERE. HELLO WE NEED HELP @OpenAI_Support
Update: It appears its the daybreak_blue flag that is broken. I changed my harness to stop sending that, and just plain GPT-5.6-Sol is giving no rejections now (even on cyber stuff it used to reject without Daybreak Blue), but if I enable that flag instant rejections.
What Happened to Daybreak Blue? Everything Worked Before, Now It’s Constantly Blocked — And OpenAI Won’t Explain Why.
I’m an approved Daybreak Blue user, and I’m extremely frustrated with what’s happening.
Until recently, everything was working fine. Now, the exact same research sessions repeatedly end with “This content can’t be shown.” Sometimes Codex refuses immediately. Other times, it works for a while, only to block everything at the end.
Hours of research, unfinished projects, and ongoing sessions are now practically useless. How are we supposed to rely on a service that suddenly stops working the way it did before?
I purchased ChatGPT Pro specifically for Daybreak Blue. That was my main reason for subscribing, and now I can barely use the capabilities I paid for.
And what’s OpenAI’s explanation? Nothing.
No clear announcement, no explanation of what changed, no indication of whether this is a temporary bug or a permanent safeguard update.
We’re left digging through community threads, trying to figure out what’s going on ourselves.
If this change is intentional, tell us. If it’s a bug, acknowledge it. But leaving paying users completely in the dark is unacceptable.
If this is how Daybreak Blue is going to operate from now on, I’ll cancel my Pro subscription and pursue Anthropic’s Claude Cyber Verification Program (CVP) instead. I see no reason to keep paying for a service that no longer supports the workflows I subscribed for.
One more thing worth investigating
In another discussion, I noticed a user pointing out that several people experiencing these issues appear to be Pro subscribers, while they personally use Plus with Daybreak Blue and haven’t experienced the same problems.
This doesn’t prove anything, but it raises an interesting question:
Could this issue be affecting Pro subscribers differently from Plus users?
I’d like to hear from others experiencing this.
OpenAI, we need answers, not another generic explanation of safety policies.
Another user pointed out:
Update: It appears its the daybreak_blue flag that is broken. I changed my harness to stop sending that, and just plain GPT-5.6-Sol is giving no rejections now (even on cyber stuff it used to reject without Daybreak Blue), but if I enable that flag instant rejections.
so it might just be this
This has not been my experience @JustinK. Lots still getting refused.
Could be a difference in what we are doing, I’m mostly doing PoCs for local code reviews.
Thanks for sharing, but that doesn’t seem to be the cause of the issue I’m experiencing. Unfortunately, the problem still persists. @Kryptium
Interesting. How do you remove that flag?
I am also doing the PoCs stuff, and it blocks me every time I ask the daybreak blue model ot do that
I don’t think so, I basically run the same prompt yesterday and it was fine. Today I am unable to even read my project at all. I can’t work, only luna is working for me, which is not much.
Then what are you working on with Daybreak? Something like Malware analysis or reverse engineering?
I’m experiencing the exact same issue. I upgraded from Plus to Pro just five days ago specifically for Daybreak Blue.
If this is a technical issue, that’s completely understandable. However, every time we contact support, we get the same generic explanations about safeguards instead of any meaningful answers. There’s no transparency, no acknowledgment of the issue, and no indication of whether it’s being investigated.
If these restrictions are intentional, I’ll cancel my recently upgraded Pro subscription and switch to CVP instead.
All we’re asking for is a clear explanation of what’s going on. Paying users shouldn’t be left in the dark about changes that directly affect the features they subscribed for.
Can confirm the same issue, it’s absolutely unusable, in fact it doesn’t want to help even with debugging for defensive work
Update: I can confirm this actually works! I switched to regular GPT-5.6 Sol without the daybreak_blue flag, and everything started working again.
Looks like this might just be a bug related to the Daybreak Blue flag rather than a broader safeguard change. Hopefully OpenAI can identify and fix it soon.
Thanks for post again.
Hey, how did you do that? @holyhonos