Hello,
This morning trying to use the new Opus 4.7 model, but getting the error message “We are unable to complete this request because it was blocked under Anthropic’s Usage Policy.” with every action I take. I had been working this project for a few weeks using Opus 4.6 with no issues. This is an internal software dev project and I can’t think of any reason this would be against policy.
I am also unable to switch back to using Opus 4.6, so I’m at a road block.
Appreciate the help.
got the same problem
random ““We are unable to complete this request because it was blocked under Anthropic’s Usage Policy.””
This is what safetyism looks like. The lockdown on Anthropic’s models will only get more strict as the models become more performant. Anthropic will certainly try to mitigate false positives, but honestly this is a very hard problem to solve. Your best option is to use models from a provider which takes a different view on the balance between safety and individual liberties.
dunno what safety im working on a preact web calculator its clearly a bug
Cursor IDE
“We are unable to complete this request because it was blocked under Anthropic’s Usage Policy.”
Every interaction with opus 4.7 in cursor ide receives this response
use opus 4.7
oh ffs
MacOS
Version: 3.0.16
VSCode Version: 1.105.1
Commit: 475871d112608994deb2e3065dfb7c6b0baa0c50
Date: 2026-04-09T05:33:51.767Z
Layout: glass
Build Type: Stable
Release Track: Default
Electron: 39.8.1
Chromium: 142.0.7444.265
Node.js: 22.22.1
V8: 14.2.231.22-electron.0
OS: Darwin arm64 25.3.0
opus 4.7
Yes - Cursor is unusable
I Have same issue:
Request ID: e82c696d-caf1-46ad-8d75-89335cc747f5
{“error”:“ERROR_OPENAI”,“details”:{“title”:“Request blocked by Anthropic”,“detail”:“We are unable to complete this request because it was blocked under Anthropic’s Usage Policy.”,“isRetryable”:false,“additionalInfo”:{},“buttons”:,“planChoices”:},“isExpected”:true}
Request blocked by Anthropic We are unable to complete this request because it was blocked under Anthropic’s Usage Policy.
NLi: Request blocked by Anthropic We are unable to complete this request because it was blocked under Anthropic’s Usage Policy.
at Bz_ (vscode-file://vscode-app/usr/share/cursor/resources/app/out/vs/workbench/workbench.desktop.main.js:28907:24552)
at Nz_ (vscode-file://vscode-app/usr/share/cursor/resources/app/out/vs/workbench/workbench.desktop.main.js:28907:23543)
at Wz_ (vscode-file://vscode-app/usr/share/cursor/resources/app/out/vs/workbench/workbench.desktop.main.js:28908:6487)
at h6u.run (vscode-file://vscode-app/usr/share/cursor/resources/app/out/vs/workbench/workbench.desktop.main.js:28908:11285)
at async vDn.runAgentLoop (vscode-file://vscode-app/usr/share/cursor/resources/app/out/vs/workbench/workbench.desktop.main.js:41216:11960)
at async zkd.streamFromAgentBackend (vscode-file://vscode-app/usr/share/cursor/resources/app/out/vs/workbench/workbench.desktop.main.js:41286:12151)
at async zkd.getAgentStreamResponse (vscode-file://vscode-app/usr/share/cursor/resources/app/out/vs/workbench/workbench.desktop.main.js:41286:18486)
at async B3e.submitChatMaybeAbortCurrent (vscode-file://vscode-app/usr/share/cursor/resources/app/out/vs/workbench/workbench.desktop.main.js:29014:16809)
at async Ma (vscode-file://vscode-app/usr/share/cursor/resources/app/out/vs/workbench/workbench.desktop.main.js:40269:4230)
@deanrie is there some way to report bugs directly to anthropic? they seems to have walled themselves into a big “we’re too important to listen to you” secret garden. the final prompt which generated this idiot error message before i gave up and switched to gpt was “hello!”. the project i was working on is a work scheduler for an industrial assembly line.
Hey, here’s some context on this error.
The message “blocked under Anthropic’s Usage Policy” comes directly from Anthropic’s own content moderation. Cursor is just passing it through, and we can’t influence the filter decision on our side. In Opus 4.7, the cyber safeguards are stricter than in 4.6, so the same project or prompt that used to work can get flagged on the newer model. Sometimes this happens even for harmless messages since the filter looks at the full chat context and the codebase.
What you can do:
@caseys, about “I can’t switch back to Opus 4.6”, can you share what exactly happens? Do you see an error, is the model missing from the list, or does the change not stick? A screenshot would help us debug it.
I think the problem is on cursor’s side.
The problem comes when agent want to ask a question or open a file like the plan. So that seems to be with the reaction cursor vs opus and not from opus alone.
hitting the same issue, 
Sharing my experience. I finally made a account in the forum just for this.
Opus 4.7 was right in blocking my request. It was badly communicated. In the project I am working, it could be interpreted as if I was trying to add negative bias to a system.
As I get more relaxed during my sessions (when everything is working), I may sometimes communicate incrementally poorly what I want and explain less why.
I was able to make it understand better why I wanted to do that, giving a example.
In my specific case, I had to provide a example that explained that I wanted to actually do the opposite. I even described a real life scenario where the ‘no-bias’ approach caused real damage.
This was enough to align with the policy, apparently.
It may look like a bad thing at first, but I noticed improved performance (on output quality) once it understood the why.
May feel like a annoying thing to deal with… however, it is a fact that if the outcome will improve the model in future training, the flagging of ‘non-compliant’ vs ‘compliant’ may actually work in our benefit: there’s clearly a layer ‘judging’ the requests, so it is clear that there’s a ‘good result’ for this judgment that is bound to influence positively the output (since ‘bad judgment’ = “request blocked mid thinking”, one can assume ‘good judgment’ = “continue thinking, this is compliant”)
TL;DR
Do not give up and explain why you are asking what you are asking, while being mindful to explain your good intentions.
Thanks Dean. User error on the 4.6 topic. I got that figured out. Regarding the safeguards on 4.6, understood that its Anthropic’s policy and they’re just passing it through cursor. They’ve obviously got some calibration to do. Re: the user who says that we need to explain our good intentions - the prompts I am talking about are so incredibly innocuous I don’t even know what I could say to provide further clarity on policy adherence.
Yeah, mine too.
But the project itself and rules (overall context as a whole) has the model thinking or generating a chain of tokens that could be predicted as “echoing credentials, keys, passwords” or engaging in a too-close-for-comfort zone where their guardrails would make sense.
I do think the guardrails currently feels off.
What I am doing in these last 2 days is tracing mentally what the model may touch or think to solve/respond to my prompt. I find it increasingly difficult to deal with when it needs to use credentials. And in my specific case in a private network, I find myself repeating often that the environment is well isolated and under control.
This is just one case.
My previous post followed another issue I had in one of the projects I help with, which touches surveillance cameras vllm-powered footage analysis.
At some point in the past a “devil’s advocate”-styled prompt was tested, and it’s goal was to use bias to fight another bias.
When creating audit tools, this specifically was triggering a lot the policy block. By revisiting the goals, showing specific files and reintroducing forgotten context pieces, I was able to craft a brief paragraph that explained the importance of it, also mentioning how the block was actually hurting my audit capability on AI outputs (which I am sure is a bigger offender) that were critical to trigger real-life humans-on-the-loop flows.
Funny enough, the prompt that got repeatedly blocked was a simple “Go ahead.” greenlighting plans and suggestions Opus 4.7 itself brought to the table in the previous message. So most of the fix was appending stuff, ‘rebuilding context’ and targeting specifically the ‘blocked by Anthropic’ issue.
Then, I felt a surge on quality (until I had to do something similar in another task early this morning).
Be careful about how you word prompts around ordinary security issues like ssh, secrets, etc - “reverse engineer” is another off limits prompt phrase
Same
Cursor IDE
When using Claude opus 4.7 in Cursor, I frequently get a “Request blocked by Anthropic’s Usage Policy” error even for completely basic, innocuous code changes. The error message reads:
“We are unable to complete this request because it was blocked under Anthropic’s Usage Policy.”
This is happening for simple tasks like:
This is happening more and more frequently and is severely impacting productivity. There is nothing remotely policy-violating about these requests.
Note: This error appears with increasing frequency even when repeating the exact same simple prompt.
The AI should process simple, benign code change requests without triggering a policy block. Basic tasks like toggling a boolean flag, updating a test assertion, or writing trivial code snippets should never be blocked by Anthropic’s Usage Policy.
MacOS
Latest (as of April 2026)
claude-opus-4-5 (opus 4.7)
90140f94-532a-420e-8201-3833af57e334
Sometimes - I can sometimes use Cursor
Me too. Many times today. Yesterday there was no problem.
I really hope Opus is not going to be same kind safety hell as GPT.
Cursor IDE
When using OPUS4.7, a prompt appears stating it cannot be used. The specific error is:We are unable to complete this request because it was blocked under Anthropic’s Usage Policy.
Request ID: 0adba916-7c9d-4342-bb9d-277f264653f1
{“error”:“ERROR_OPENAI”,“details”:{“title”:“Request blocked by Anthropic”,“detail”:“We are unable to complete this request because it was blocked under Anthropic’s Usage Policy.”,“isRetryable”:false,“additionalInfo”:{},“buttons”:,“planChoices”:},“isExpected”:true}
Request blocked by Anthropic We are unable to complete this request because it was blocked under Anthropic’s Usage Policy.
YLi: Request blocked by Anthropic We are unable to complete this request because it was blocked under Anthropic’s Usage Policy.
at rK_ (vscode-file://vscode-app/d:/Agents/cursor/resources/app/out/vs/workbench/workbench.desktop.main.js:28926:24552)
at nK_ (vscode-file://vscode-app/d:/Agents/cursor/resources/app/out/vs/workbench/workbench.desktop.main.js:28926:23543)
at dK_ (vscode-file://vscode-app/d:/Agents/cursor/resources/app/out/vs/workbench/workbench.desktop.main.js:28927:6487)
at h8u.run (vscode-file://vscode-app/d:/Agents/cursor/resources/app/out/vs/workbench/workbench.desktop.main.js:28927:11285)
at async xDn.runAgentLoop (vscode-file://vscode-app/d:/Agents/cursor/resources/app/out/vs/workbench/workbench.desktop.main.js:41235:11960)
at async W1d.streamFromAgentBackend (vscode-file://vscode-app/d:/Agents/cursor/resources/app/out/vs/workbench/workbench.desktop.main.js:41305:12151)
at async W1d.getAgentStreamResponse (vscode-file://vscode-app/d:/Agents/cursor/resources/app/out/vs/workbench/workbench.desktop.main.js:41305:18486)
at async N3e.submitChatMaybeAbortCurrent (vscode-file://vscode-app/d:/Agents/cursor/resources/app/out/vs/workbench/workbench.desktop.main.js:29033:16809)
at async Va (vscode-file://vscode-app/d:/Agents/cursor/resources/app/out/vs/workbench/workbench.desktop.main.js:40288:4233)
Suggest policy issues?
Windows 10/11
Version: 3.1.17 (user setup)
VSCode Version: 1.105.1
Commit: fce1e9ab7844f9ea35793da01e634aa7e50bce90
Date: 2026-04-19T19:33:58.189Z
Layout: editor
Build Type: Stable
Release Track: Default
Electron: 39.8.1
Chromium: 142.0.7444.265
Node.js: 22.22.1
V8: 14.2.231.22-electron.0
OS: Windows_NT x64 10.0.26200
Yes - Cursor is unusable
@Serp atm it seems worse! Hopefully it’s just a glitch, or maybe they are trying to block Cursor to use their model? So that people switch to Claude? Who knows.
Claude alone is no match for Cursor. Everything is tiny little bit not so good done there. Good addition when credits run out, but Cursor is still the best for coding.
But I have started using Claude for chat instead openai. It feels faster and able.
