Skip to content

When ChatGPT declines security work

There is no uncensored ChatGPT: it is one company’s hosted assistant, and its refusal behaviour cannot be switched off by a user. If it declines security questions on an engagement you are authorized to run, the practical alternative is a different model, and PentestLLM serves 9 security-tuned builds by API for that.

Get a keyPrices per token
python
from openai import OpenAI

client = OpenAI(base_url="https://api.pentestllm.ai/v1", api_key="sk-ptl-live-…")

reply = client.chat.completions.create(
    model="auto/security",
    messages=[{"role": "user", "content": "Explain this advisory and list what to check on an affected host."}],
)
print(reply.model)  # the build that answered
print(reply.choices[0].message.content)
Rows of server cabinets in a data centre

Why a general assistant declines

A consumer assistant serves a very large public audience and has no way to check who is asking. It cannot see your statement of work, so it treats a question about a vulnerability the same whether it comes from a tester or from someone with no right to ask. Declining broadly is a sensible design for that product. It is a poor fit for a job where the questions are offensive by nature and the authorization lives in a contract the assistant never sees.

The result is familiar to anyone who has tried it mid-engagement: a refusal paragraph, a lecture, or an answer so hedged it is no use, in response to a routine question.

Why a workaround is the wrong fix

People search for chatgpt uncensored hoping for a setting or a special prompt. There is no setting. Prompts that try to argue the assistant out of its training work inconsistently, stop working when the model is updated, and put the account you use for client work on the wrong side of its terms. Jailbreak prompts vs abliterated models explains why that route is unstable.

The durable fix is to use a model that was built to answer security questions in the first place.

What changes when you switch

Refusals
Security-tuned builds answer the field’s routine questions. Refusals are reduced, not zero.
Interface
ChatGPT is a chat window. This is an API that plugs into a terminal agent, an editor or a chat front end you choose.
Billing
A subscription charges monthly. Here you pay per token from a prepaid balance, from $10 by card.
Model choice
One assistant becomes 9 security builds plus the abliterated builds, switched by changing one id.
Extras
There is no built-in memory, voice or image generation. You get models, and the tools the API runs for them.

If you still want a chat window

An API does not have to mean a terminal. Any chat client that accepts an OpenAI-compatible base URL can sit in front of it: enter https://api.pentestllm.ai/v1, paste the key and type auto/security as the model. Editors and agents work the same way, and the agent guides list the fields for each. For hosted assistants with their own interface, the HackerGPT alternative page compares the options.

The API can run web search and page fetch on its side for builds that take tool calls, so a model asked about a recent advisory can read it instead of guessing. The fees are on the pricing page.

Habits to keep and habits to drop

Keep the habit of giving context: say what the engagement is, what is in scope and what format you want back. Drop the habit of padding a request with justifications to avoid a refusal, because a security build does not need them and they cost input tokens. Keep the habit of checking. A model that answers readily is still a model, and its claims about versions, flags and behaviour need verifying before they reach a report.

Authorized use only

Switching models does not change what you may test. PentestLLM is for lawful security work on systems you own or have written permission to test: client engagements, bug bounty programs you are enrolled in, CTF events and your own lab. The acceptable use policy has the full statement, and uncensored AI describes what the word covers on this site.

9 security builds, priced per token

Read from the live catalog on 2026-11-12. Prices are US dollars per million tokens, paid from a prepaid balance that starts at $10. The id auto/security has no price of its own: each request is billed at the build that answered.

BuildInputOutput
Qwen3.8 27B Cybersecurity$0.40$2.40
Gemma 4 26B A4B Cybersecurity$0.4224$1.3376
GLM-5.3 Flash Cybersecurity$0.5093$1.6975
Red Team (Flash)$0.90$3.00
Red Team (Pro)$1.50$6.00
OrcaCyber MiniMax M3.1 Flash$2.00$4.00
OrcaCyber Qwen3.8 Flash Next$2.00$4.00
Cyber Frontier$5.25$15.75
OrcaCyber Zero 1.5$7.50$18.75

Full price list and tool fees · Every model · API docs

Questions

  • No. ChatGPT is a hosted assistant and its refusal behaviour is set by its maker. What exists are other models, tuned or edited to decline less, that you can call instead.

  • A security-tuned model reached by API. PentestLLM serves 9 of them on one prepaid key, and the model id auto/security picks one per request.

  • Far more often, because the builds are tuned or abliterated for security work. They can still decline sometimes, so the honest claim is reduced refusals.

  • Yes, through any chat client that accepts an OpenAI-compatible base URL and key. PentestLLM itself is an API, not a chat app.

  • No. You top up a prepaid balance from $10 by card and pay per token. On 2026-11-12 the lowest-priced security build costs $0.40 per million input tokens.

Read next

One key, from $10

PentestLLM is for systems you own or have written permission to test. The acceptable use policy says what is not allowed.

Get a key

usage, 1,000 agent turns on Qwen3.8 27B Cybersecurity
Input, 8,000 tokens at $0.40 per million
$0.0032
Output, 1,000 tokens at $2.40 per million
$0.0024
One turn
$0.0056
1,000 turns
$5.60
output price per million tokens, 9 builds

$1.3376 Gemma 4 26B A4B CybersecurityOrcaCyber Zero 1.5 $18.75